JP2004289288A - Service list selecting apparatus and method, and program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable a network wide model to optimize a service list and optimize the assignment state of each logical path. <P>SOLUTION: The service list selecting apparatus acquires a call generating rate λ<SP>s(i, e)</SP>, a call end rate μ<SP>s(i, e)</SP>, and a selection probability formula P<SP>s(i, e)</SP>with respect to a network end node pair (i, e), acquires a network topology, band information of each link, constraint conditions with respect to a bandwidth object and a path object of each logic path, a reference of a unit service network resource amount function r<SB>j</SB>(SL) and a set of service list objects, calculates the selection probability formula P<SP>s(i, e)</SP>and calculates the unit service network resource amount function r<SB>j</SB>(SL). The apparatus uses calculation results of the λ<SP>s(i, e)</SP>, the μ<SP>s(i, e)</SP>, the P<SP>s(i, e)</SP>, r<SB>j</SB>(SL) to decide a logic path object path<SP>(i, e)</SP>in matching with the constraint conditions, calculates an expected income obtained when each of the path<SP>(i, e)</SP>is selected as a target value, and calculates an optimum target function value and optimum logical path assignment. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

α_ｎ：重みベクトルαの制御区間ｎにおける推定値
Ｐ_ｎ：α_ｎの共分散行列（ｍ×ｍ次元）
Ｋ_ｎ：カルマンゲイン（ｍ次元ベクトル）
Ｑ_ｎ：システムノイズの共分散行列（ｍ×ｍ次元）
また、実用的にはＱ_ｎとしてｎに無関係な正定値行列（例Ｑ_ｎ＝０．０５Ｉ）を用い、初期値としてα_０はゼロベクトル、（Ｐ_０＝１０００Ｉ）を用いればよい。
【００５３】
この例では、需要予測機能部２３は、制御区間ｎの終了時に、観測された呼発生個数ｙ^ｓ _ｎを履歴データベース２１から読み出して、第（１−３）式〜第（１−６）式を用いて、α_ｎを更新する。また、制御区間ｎをインクリメントして新たな制御区間ｎの開始時に、第（１−１）式によってｙ^ｓ＾ _ｎを算出する。
制御区間の開始時に制御を集中させると、制御区間（ｎ−１）が終了して制御区間ｎが開始する時点で、次のような処理を行うことになる。まず、観測された呼発生個数ｙ^ｓ _ｎ−１を履歴データベース２１から読み出して、次に示す第（２−３）式〜第（２−６）式によりα_ｎ−１を更新する。 そして、更新されたα_ｎ−１を用い第（１−１）式によってｙ^ｓ＾ _ｎを算出する。
【数２】

呼平均継続時間ｈ^ｓ＾ _ｎの予測については、ｙ^ｓ _ｎをｈ^ｓ _ｎに置き換えることで、上記と同様に逐次的にα_ｎ−１を更新して、その更新されたα_ｎ−１を用いて、ｈ^ｓ＾ _ｎを算出することで実現できる。
需要予測機能部２３は、１つの網端ノードｉともう１つの網端ノードｅとのペア（ｉ，ｅ）についてそのノード間に関する需要予測を行う。
【００５４】
図１に示す最適サービスリスト決定機能部１０は、図３に示す処理を行う。
図３のステップＳ０１では、最適サービスリスト決定機能部１０の情報取得部１１は、需要予測機能部２３から各網端ノードペア（ｉ，ｅ）に関する予測呼発生率λ^{ｓ（ｉ，ｅ）}及び予測呼終了率μ^{ｓ（ｉ，ｅ）}を取得する。
ステップＳ０２では、最適サービスリスト決定機能部１０の情報取得部１２は、選択確率モデル化機能部２４から、セグメントｓがサービスリストＳＬからサービスクラスｊを選択する確率を表す選択確率式Ｐ^ｓ（ｊ，ＳＬ）を取得する。
【００５５】
ステップＳ０３では、最適サービスリスト決定機能部１０の情報取得部１３は、計算条件データベース２２から次の情報を取得する。
（１）網トポロジ
（２）各リンクの使用可能な帯域幅（例えば、ノードｎ１とノードｎ２との間に存在する接続リンクの使用可能帯域幅を各要素とする行列Ｒ（ｎ１，ｎ２））
各論理パスの帯域幅候補と経路候補に関する制約条件
（３）入力されるあるサービスリストＳＬの各サービスクラスｊの１ユーザあたりに必要とされる網資源量を表す単位サービス網資源量関数ｒ_ｊ（ＳＬ）のレファレンス
（４）ＳＬとして取りうる候補の集合
なお、ステップＳ０１，Ｓ０２，Ｓ０３についてはいずれの順番で順次に実行してもよいし、同時に実行してもよい。
【００５６】
ステップＳ０４では、前のステップＳ０１〜Ｓ０３で取得した情報に基づいて、目標値最大化サービスリスト選出部１８がＳＬの全候補の中から目標値が最大となるサービスリスト及びその際の最適論理パス割り当て状態を選び出す。
【００５７】
実際には、ステップＳ０４は最適サービスリスト決定機能部１０の処理によって実現される。最適サービスリスト決定機能部１０の動作の概要は、図４に示す通りである。以下、図４の動作について説明する。
ステップＳ１０では、目標値の最大値を格納するための変数Ｍａｘを０に初期化する。
【００５８】
情報取得部１３が計算条件データベース２２から取得したサービスリストＳＬの候補のそれぞれに対して、ステップＳ１１以降の処理が順次に実行される。ステップＳ１１では、未処理のサービスリストＳＬの候補が残っているか否かを調べる。
未処理のサービスリストＳＬの候補が存在する場合には、ステップＳ１１からＳ１２に進む。存在しなければ処理を終了する。
【００５９】
ステップＳ１２では、未処理の候補群の中から１つのサービスリストＳＬの候補を取り出し、それに含まれているサービスクラスの数、すなわちクラス数Ｊを認識する。
次のステップＳ１３では、単位サービス網資源必要量算出部１５を用いて、選択されたサービスリストＳＬの候補に含まれている全ての要素（サービスクラス）について、サービスクラスｊ毎に単位サービス網資源必要量ｒ_ｊ（ＳＬ）を算出する。
【００６０】
また、ステップＳ１４では選択確率算出部１４を用いて、全てのセグメントｓ及び全てのサービスクラスｊについて、選択確率式Ｐｓ（ｊ，ＳＬ）を算出する。
なお、ステップＳ１３，Ｓ１４については何れか一方を先に実行してもよいし、同時に実行してもよい。
次のステップＳ１５では、論理パス目標関数値算出部１６を用いて、各網端ノードペア（ｉ，ｅ）間に割り当てられる論理パスについて、需要予測機能部２３から取得した予測呼発生率λ^{ｓ（ｉ，ｅ）}及び予測呼終了率μ^{ｓ（ｉ，ｅ）}と、選択確率算出部１４で算出された選択確率式Ｐ^ｓ（ｊ，ＳＬ）と、単位サービス網資源必要量算出部１５で算出された単位サービス網資源必要量ｒ_ｊ（ＳＬ）とに基づき、計算条件データベース２２から取得した各論理パスの帯域幅候補及び経路候補に関する制約条件に合致する有限個の離散値設定帯域幅候補と、有限個の経路候補の組み合わせで構成される有限個の論理パス候補集合Ｐａｔｈ^{（ｉ，ｅ）}を設定し、Ｐａｔｈ^{（ｉ，ｅ）}の各要素について、仮にそれが選択された場合に予想される各々の論理パスで得られる予想収入を目標値Ｔａｒｇｅｔ（ＳＬ，（ｉ，ｅ，ｒ，ｂ））として算出する。
【００６１】
ここで、ｒは経路候補のＩＤ（識別子）を表し、ｂは経路候補の中でのＩＤを表す。すなわち、（ｉ，ｅ，ｒ，ｂ）で論理パス候補集合Ｐａｔｈ^{（ｉ，ｅ）}の１要素を特定する。
次のステップＳ１６では、最適論理パス選択部１７を用い、各経路が選択された場合には「１」、選択されない場合には「０」となる０−１変数を全ノード間の全論理パス分用意して、それと論理パス目標関数値算出部１６で算出された各論理パス候補の目標値Ｔａｒｇｅｔ（ＳＬ，（ｉ，ｅ，ｒ，ｂ））を用いて、各リンクを通過する選択される論理パス設定帯域の合計が使用可能リンク帯域を超えないという制約条件と、論理パス候補のうち選択されるのは１つであるという制約条件との下で、各論理パスから得られる合計収入を目標関数値ＴａｒｇｅｔＶａｌｕｅとし、またその最大化を最適化条件とする０−１整数計画問題として定式化する。その際に、目標関数値ＴａｒｇｅｔＶａｌｕｅをＭａｘに初期化する。
【００６２】
ステップＳ１７では、ステップＳ１６で定式化された０−１整数計画問題を最適化するように解き、目標関数値ＴａｒｇｅｔＶａｌｕｅを求める。
ステップＳ１８では、結果として得られた目標関数値ＴａｒｇｅｔＶａｌｕｅとＭａｘとを比較する。（ＴａｒｇｅｔＶａｌｕｅ＞Ｍａｘ）の条件を満たす場合に限り次のステップＳ１９に進む。
【００６３】
ステップＳ１９では、ＭａｘをＴａｒｇｅｔＶａｌｕｅに更新する。また、そのときのサービスリストＳＬと各論理パス割り当て状態（経路と帯域幅）をそれぞれ最適サービスリストＳＬ及び最適論理パス割り当てとして保存する。
ステップＳ１１以降の処理を全てのサービスリストＳＬの候補について順次に実行すると、その結果として最終的に最適サービスリストＳＬ及び最適論理パス割り当てが算出される。
【００６４】
最適論理パス選択部１７においては、０−１整数計画問題を以下に示すようにして定式化する。
まず、与えられる条件である物理的網に関して、次のパラメータを定義する。Ｃ（ｎ１，ｎ２）：リンク（ｎ１，ｎ２）が存在するならその使用可能帯域幅、存在しなければ０を割り当てる。なお、リンク（ｎ１，ｎ２）とはノードｎ１とノードｎ２とを直結するリンクを意味する。また、リンク集合を（ＬＩＮＫ）とする。
【００６５】
Ａｌｌ：論理パス候補の全集合。各論理パスのＩＤは（ｉ，ｅ，ｒ，ｂ）であり、網端ノードｉと網端ノードｅとの間に設定される論理パス候補のうち（経路候補ＩＤ＝ｒ，帯域幅候補ＩＤ＝ｂ）のものを表す。
ｎｕｍＲ_{（ｉ，ｅ）}：論理パス（ｉ，ｅ）の経路候補数，ＩＤはｒ。
ｎｕｍＢ_{（ｉ，ｅ）}：論理パス（ｉ，ｅ）の帯域幅候補数，ＩＤはｂ。
【００６６】
ｇ^{（ｉ，ｅ）ｒ} _{（ｎ１，ｎ２）}：論理パス（ｉ，ｅ）の経路候補ＩＤ＝ｒがリンク（ｎ１，ｎ２）を通過するなら１、それ以外なら０となる。
ＢＷ^{（ｉ，ｅ）ｂ}：論理パス（ｉ，ｅ）の帯域幅候補ＩＤ＝ｂの割り当て帯域幅。
ｉｎｃｏｍｅ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ）：サービスリストをＳＬに定め、論理パス（ｉ，ｅ，ｒ，ｂ）が定められた場合にこの論理パスから得られる予想収入値となる。
【００６７】
ｔａｒｇｅｔ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ）：サービスリストをＳＬに定め、論理パス（ｉ，ｅ，ｒ，ｂ）が定められた場合にこの論理パスから得られる目標関数値。収入を目標とする場合には（ｉｎｃｏｍｅ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ））と同じになる。
ｘ^{（ｒ，ｂ）} _{（ｉ，ｅ）}：論理パス（ｉ，ｅ）が経路として（候補ＩＤ＝ｒ）を、割り当て帯域幅として（候補ＩＤ＝ｂ）をとる場合に「１」、それ以外の場合に「０」となる０−１変数。これが求めるべき解となる。
【００６８】
定式化に必要なパラメータは以上の通りである。最適化関数及び目標は、ｅ２ｅ論理パスで得られる収入期待値の合計の最大化なので次式の通りとなる。
【数３】

前記第（３−２）式は、（ｉ，ｅ）の論理パスに関する経路と割り当て帯域幅の両方を考慮に入れたＬＳＰ候補のうち、どれか１つのみを選択しうるという排他条件を表している。また、第（３−３）式は各ｅ２ｅ論理パスがそれぞれの論理パス候補を選択した場合に（ｎ１，ｎ２）のリンクにおいて各論理パスが要求する帯域幅合計値が、使用可能リンク帯域より小さいことを意味する制約条件を表す。
【００６９】
前記第（３−１）式〜第（３−４）式において、全ての（ｉ，ｅ）及び全てのｒ及び全てのｂに関するｇ^{（ｉ，ｅ）ｒ} _{（ｎ１，ｎ２）}とＢＷ^{（ｉ，ｅ）ｂ}とその場合のｔａｒｇｅｔ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ）が決まれば、変数はｘ^{（ｒ，ｂ）} _{（ｉ，ｅ）}のみの０−１整数計画問題となる。これは、分岐限定方法などの既存の解法により解くことが可能である。
この形態のサービスリスト算出装置では、論理パス目標関数値算出部１６が読み出す各論理パスの帯域幅候補と経路候補に関する制約条件の具体例として、次のものを想定している。
【００７０】
論理パス経路候補の条件の例：
［経路候補条件（１）］ホップ数が「最短経路ホップ数＋α（与えられた一定正整数）」以下である経路。
［経路候補条件（２）］合計リンク遅延が「最小合計リンク遅延＋β（与えられた一定正整数）」以下である経路。
【００７１】
論理パス帯域幅候補の条件の例：
［帯域幅候補条件（１）］予め各論理パス毎にもしくは全論理パスに対して共通で用意された有限個数の固定帯域幅を候補とする。
【００７２】
［帯域幅候補条件（２）］論理パスの設定可能な最小単位をＢｗＵｎｉｔとして決め、予め各論理パス毎にもしくは全論理パスに対して共通で用意された有限個数の目標呼損率それぞれに対して、予想呼損率が目標呼損率以下となる最大の帯域幅を帯域幅候補とする。サービスリストが変わる毎に動的に各目標呼損率を満たす最大帯域幅を算出する。
【００７３】
［帯域幅候補条件（３）］論理パスの設定可能な単位をＢｗＳｔｅｐとして決め、予め各論理パス毎にもしくは全論理パスに対して共通で用意された１つの目標呼損率に対して、予想呼損率が目標呼損率以下となる最大の帯域幅を基準帯域幅とし、予め決められた一定個数ずつ基準帯域幅の上下に設定可能な離散帯域幅を含めて、帯域幅候補とする。サービスリストが変わる毎に動的に基準帯域幅とその上限の離散帯域幅を算出する。
【００７４】
ここでは、前記［経路候補条件（１）］と［帯域幅候補条件（１）］とを適用する場合に、呼の生起がランダム生起、継続時間が指数時間に従う場合を想定して、あるサービスリストＳＬが割り当てられた場合の、論理パス目標関数値算出部１６での（Ｔａｒｇｅｔ（ＳＬ，ｐａｔｈ^{（ｉ，ｅ）}））の計算式を示す。
まず、ある（ｉ，ｅ）を固定する。論理パス目標関数値算出部１６においては、［経路候補条件（１）］に合う経路を全て抽出する。ノードｉから順にホップ数を１つずつ増やしてループをしない木構造の接続関係図を作り、最短経路を算出した後、ホップ数が［最短経路＋α］に届くまで、もしくはノードｅに届くまで続けて、伸ばす枝がなくなったら、ノードｅまで届いた枝に関してのみ、枝を逆にたどると経路候補ができあがる。［帯域幅候補条件（１）］は読み出された各帯域幅をそれぞれ帯域幅ＢＷ^{（ｉ，ｅ）ｂ}に設定するのみである。
【００７５】
次に（Ｔａｒｇｅｔ（ｉ，ｅ，ｒ，ｂ，ＳＬ））の計算について説明する。収入を目標値（Ｔａｒｇｅｔ（ｉ，ｅ，ｒ，ｂ，ＳＬ））とする場合には、経路による違いは発生しないため、設定された帯域幅ＢＷ^{（ｉ，ｅ）ｂ}の１ボトルネックモデルとみなすことができる。
【００７６】
多呼源トラヒックモデルを用いた場合について説明する。
選択確率算出部１４で算出された選択確率式Ｐ^ｓ（ｊ，ＳＬ）を使って、ＳＬの場合の（ｉ，ｅ）間に発生するサービスクラスｊに対する生起呼量ａ_ｊ ^{（ｉ，ｅ）}を次式から算出する。
ａ_ｊ ^{（ｉ，ｅ）}＝Σ_ｓ｛λ^{ｓ（ｉ，ｅ）}×Ｐ^ｓ（ｊ，ＳＬ）／μ^{ｓ（ｉ，ｅ）}｝ ・・・（４−１）
次に、Ｐ^ｓ（ｊ，ＳＬ）と、ａ_ｊ ^{（ｉ，ｅ）}と、ＢＷ^{（ｉ，ｅ）ｂ}と、単位サービス網資源必要量算出部１５が算出したｒ_ｊ（ＳＬ）とを使い、（０≦Σ_ｊ（ｎ_ｊ×ｒ_ｊ（ＳＬ））≦ＢＷ^{（ｉ，ｅ）ｂ}）の条件を満たす全ての定常状態確率ｐ（ｎ１，・・・，ｎｊ，・・・，ｎＪ）を次式で算出する。
【００７７】
ｐ（ｎ１，・・・，ｎｊ，・・・，ｎＪ）＝Π^Ｊ _ｊ（（ａ_ｊ）^ｎｊ／ｎ_ｊ！）×ｐ（０，０，，，，０） ・・・（４−２）
但し、
ｐ（０，．．．０）＝１／［Σ^ＮＪ _ｎｊ＝０・・・Σ^Ｎ２ _ｎ２＝０Σ^Ｎ１ _ｎ１＝０Π^Ｊ _ｊ（（ａ_ｊ）^ｎｊ／ｎ_ｊ！）］，
Ｎ_ｊ＝ＩＮＴ［ＢＷ^{（ｉ，ｅ）ｂ}−Σ^Ｊ _{ｋ＝ｊ＋１}（ｎ_ｋ×ｒ_ｋ（ＳＬ）））／ｒ_ｊ（ＳＬ））］，
ＩＮＴ［ａ］はａの小数点以下を切り捨てる整数化を表す。
【００７８】
そして、この論理パス候補ｐａｔｈ^{（ｉ，ｅ）}から予測収入値を次式から求める。

但し、
Ｎ_ｊ＝ＩＮＴ［ＢＷ^{（ｉ，ｅ）ｂ}−Σ^Ｊ _{ｋ＝ｊ＋１}（ｎ_ｋ×ｒ_ｋ（ＳＬ））／ｒ_ｊ（ＳＬ）］，
ｐｒｉｃｅ_ｊはサービスリストＳＬに含まれているクラスｊの単位時間当たり価格。
【００７９】
また、単呼源モデルとしてモデル化することも可能である。まず、（ｉ，ｅ）間における各セグメントｓの生起呼量ａ^ｓと聡生起呼量Ａを次式から算出する。
ａ^{ｓ（ｉ，ｅ）}＝λ^{ｓ（ｉ，ｅ）}／μ^{ｓ（ｉ，ｅ）}，Ａ^{（ｉ，ｅ）}＝Σ_ｓａ^{ｓ（ｉ，ｅ）} ・・・（４−４）
また、最大状態数Ｎを次式で算出する。

この式の意味について説明する。今、（ｉ，ｅ）間における全セグメント合計分の呼の状態数がｎである状況を想定する。ここで、セグメントｓからの呼の数は生起呼量の比よりも平均的に（（ａ^ｓ／Ａ）×ｎ）であると見積もることができ、更にセグメントｓの呼がサービスクラスｊを選ぶ呼の数は（（ａ^ｓ／Ａ）×Ｐ^ｓ（ｊ，ＳＬ）×ｎ）と見積もることができる。
【００８０】
サービスクラスｊを選ぶ呼は、網資源ｒ_ｊ（ＳＬ）を必要とするため、全セグメントの全サービスクラスでは合計で（Σ_ｊ（Σ_ｓ［ａ^ｓ／Ａ×Ｐ^ｓ（ｊ，ＳＬ）］×ｒ_ｊ（ＳＬ））×ｎ）の網資源が必要になると考えられる。これがＢＷ^{（ｉ，ｅ）ｂ}以下ならば収容できるということなので、次の関係式が成立する。
Σ_ｊ（Σ_ｓ［ａ^ｓ／Ａ×Ｐ^ｓ（ｊ，ＳＬ）］×ｒ_ｊ（ＳＬ））×ｎ≦ＢＷ^{（ｉ，ｅ）ｂ} ・・・（４−６）
よって、ｎとして取りうる最大状態数Ｎは次のように見積もることができる。
【００８１】
Ｎ＝ＩＮＴ［ＢＷ^{（ｉ，ｅ）ｂ}／Σ_ｊ（Σ_ｓ［ａ^ｓ／Ａ×Ｐ^ｓ（ｊ，ＳＬ）］×ｒ_ｊ（ＳＬ））］ ・・・（４−７）
すると、状態ｎ（１≦ｎ≦Ｎ）に関する定常状態確率ｐ（ｎ）は次式で表される。
ｐ（ｎ）＝（Ａ^ｎ／ｎ！）×ｐ（０）
ｐ（０）＝１／［１＋Σ^Ｎ _ｎ（Ａ^ｎ／ｎ！）］
すなわち、単一待ち行列モデルＭ／Ｍ／Ｎ（０）での定常状態確率式で算出することができる。目標関数は、この（ｉ，ｅ，ｒ，ｂ）から得られる予想収入なので、次式から求められる。
【００８２】

この式について説明する。
前述の場合と同様に、全セグメント分の呼の状態数がｎである状況を想定する。その際は、そのうちセグメントｓからの呼の数は生起呼量の比より（（ａｓ／Ａ）×ｎ）と平均的に見積もることができ、更にセグメントｓの呼がサービスクラスｊを選ぶ呼の数は（（ａ^ｓ／Ａ）×Ｐ^ｓ（ｊ，ＳＬ）×ｎ）と見積もることができる。
【００８３】
よって、全セグメントｓに関してサービスクラスｊを選ぶ呼の合計数は（Σ_ｓ｛ｎ×（ａ^ｓ／Ａ）×Ｐ^ｓ（ｊ，ＳＬ）｝）と見積もることができるため、状態ｎで全呼から得られる収入は（Σ_ｊ［ｐｒｉｃｅ_ｊ×（Σ_ｓ［ｎ×（ａ^ｓ／Ａ）×Ｐ^ｓ（ｊ，ＳＬ）］）］）で算出できる。
これに状態確率ｐ（ｎ）を乗算して全ての状態に関して加算するので、これは網収入の期待値を意味する。
【００８４】
多呼源待ち行列モデルによるｔａｒｇｅｔ（ｉ，ｅ，ｒ，ｂ，ＳＬ）の計算と、単一待ち行列モデルによる計算とを比較する。前者は、どんな状況でも計算精度が高いが、計算量が膨大になる。後者は、計算量が少なく済むが、一般的に計算精度が大きく劣化する。ただ、後者は呼損が殆ど発生しないような状況では、多呼源モデルと変わらない高い計算精度となるため、本制御の対象となるトラヒックが使用できる帯域がサービス選択の結果要求される帯域幅よりも圧倒的に多い状況において利用することが適切であると考えられる。
【００８５】
以上のように、ネットワークワイドモデルである通信ネットワークにおいて、複数のサービスクラスを提供する場合に、各網端ノード間の需要とサービスリストによるユーザのサービス選択確率変化と論理パス割り当て状態を考慮に入れて、網収入を最大化するサービスリスト及びその際の論理パス割り当て状態を決定することができる。
【００８６】
なお、図１に示すサービスリスト算出装置においては、各網端ノードペア（ｉ，ｅ）に関する呼発生率λ^{ｓ（ｉ，ｅ）}及び呼終了率μ^{ｓ（ｉ，ｅ）}並びに選択確率式Ｐ^ｓ（ｊ，ＳＬ）を需要予測機能部２３，選択確率モデル化機能部２４で履歴情報から予測しているが、予め計算などにより求めた呼発生率λ^{ｓ（ｉ，ｅ）}及び呼終了率μ^{ｓ（ｉ，ｅ）}並びに選択確率式Ｐ^ｓ（ｊ，ＳＬ）を定数としてデータベースに保持しておけは、需要予測機能部２３及び選択確率モデル化機能部２４を省略して図５に示すように構成を簡略化することもできる。
【００８７】
（第２の実施の形態）
本発明のサービスリスト選択装置及び方法並びにプログラム及び記録媒体のもう１つの実施の形態について、図６を参照して説明する。この形態は、請求項２，請求項５，請求項８に対応する。
図６はこの形態のサービスリスト算出装置の構成を示すブロック図である。この形態は、第１の実施の形態の変形例である。図６において図１と対応する要素は同一の符号を付けて示してある。第１の実施の形態と同一の部分については、以下の説明を省略する。
【００８８】
図６に示すように、この形態のサービスリスト算出装置も履歴データベース２１，計算条件データベース２２Ｂ，需要予測機能部２３，選択確率モデル化機能部２４及び最適サービスリスト決定機能部１０を備えている。
計算条件データベース２２Ｂに保持される情報として、網の各リンク毎のコスト情報（各リンクの単位帯域当たりの費用を表す）が追加されている。この形態の最適サービスリスト決定機能部１０は、このリンク毎のコスト情報も用いて計算を行う。
【００８９】
最適サービスリスト決定機能部１０の動作の概略は図３と同様であるが、ステップＳ０３では計算条件データベース２２Ｂからリンク毎のコスト情報も取得する。
また、ステップＳ０４の「目標値最大化サービスリスト選出」の内容は図４と同様である。
【００９０】
ここで、第１の実施の形態の場合と同様に、前記［経路候補条件（１）］と［帯域幅候補条件（１）］とを適用する場合に、呼の生起がランダム生起、継続時間が指数時間に従う場合を想定して、あるサービスリストＳＬが割り当てられた場合の、論理パス目標関数値算出部１６での収益に関する（Ｔａｒｇｅｔ（ｉ，ｅ，ｒ，ｂ，ＳＬ））の計算式を示す。
【００９１】
ここで、パラメータとして以下のものを定義する。
Ｃｏｓｔ（ｎ１，ｎ２）：リンク（ｎ１，ｎ２）の単位時間・単位帯域当たりのコスト。
Ｉｎｃｏｍｅ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ）：サービスリストをＳＬに設定され、且つ論理パス（ｉ，ｅ，ｒ，ｂ）が選択された場合にこの論理パスから得られる予想収入値。
Ｅｘｐｅｎｄｉｔｕｒｅ（ｉ，ｅ）（ｒ，ｂ，ＳＬ）：サービスリストをＳＬに設定され、且つ論理パス（ｉ，ｅ，ｒ，ｂ）が選択された場合にこの論理パスの設定帯域及びその経路にかかる予想費用値。
【００９２】
この場合、予想費用値Ｅｘｐｅｎｄｉｔｕｒｅ（ｉ，ｅ）（ｒ，ｂ，ＳＬ）は次式で求められる。

また、予想収入値Ｉｎｃｏｍｅ_{（ｉ，ｅ）}（ｒ，ｂ，ＳＬ）は前述の通りであるので、多呼源トラヒックモデルを用いた場合は、収益（ｔａｒｇｅｔ（ｉ，ｅ）（ｒ，ｂ，ＳＬ））は次式で表される。
【００９３】

但し、
Ｎ_ｊ＝ＩＮＴ［ＢＷ^{（ｉ，ｅ）ｂ}−Σ^Ｊ _{ｋ＝ｊ＋１}（ｎ_ｋ×ｒ_ｋ（ＳＬ））／ｒ_ｊ（ＳＬ）］，
ｐｒｉｃｅ_ｊはサービスリストＳＬに含まれているクラスｊの単位時間当たり価格。
【００９４】
また、単一呼源待ち行列モデル適用した場合には、収益（ｔａｒｇｅｔ（ｉ，ｅ）（ｒ，ｂ，ＳＬ））は次式で表される。

以上のように、ネットワークワイドモデルのある通信ネットワークにおいて、複数サービスクラスを提供する場合に、各網端ノード間の需要とサービスリストによるユーザのサービス選択確率変化と、論理パス割り当て状態を考慮に入れて、網収益を最大化するサービスリスト及びその際の論理パス割り当て状態を決定することができる。
【００９５】
なお、本発明のサービスリスト選択方法によって算出された最適なサービスリスト及び最適な論理パス割り当て状態を直ちに次の制御期間に適用することも可能ではあるが、それが望ましいとは限らない。
例えば、ユーザから見た場合、次の制御期間のサービスリスト（料金体系）が未定であるのはサービスとして不安に感じる可能性もある。従って、実際の通信サービスにおいては、適用する予定のサービスリストを一定期間（例えば１週間や１ヶ月）前もって公示しておくようなサービスを実施することも考えられる。
【００９６】
そのような場合には、本発明を実施するシステムはサービスリスト提案システムとしてバックグラウンドで稼働させておき、網サービス提供事業者の判断材料として算出されたサービスリストの情報を利用することが考えられる。
また、論理パス割り当て制御に関しても、本発明の方法により算出される最適な論理パス割り当て状態を制御期間で固定として制御することも可能であるが、実際のトラヒック量の変動に対して、論理パスの再割り当て制御（帯域の増減など）を行うことも技術的には可能である。
【００９７】
このような現実の実施状況については、環境や必要性に応じて適宜変更すればよい。
【００９８】
【発明の効果】
以上述べたように、本発明によれば、ネットワークワイドモデルに相当する通信ネットワークにおいて、複数サービスクラスを提供する場合に、各網端ノード間の需要とサービスリストによるユーザのサービス選択確率変化と論理パス割当状態を考慮にいれて、網収入や網収益を最大化する最適サービスリスト及び最適論理パス割当状態を決定することが可能となる。
【図面の簡単な説明】
【図１】第１の実施の形態のサービスリスト算出装置の構成を示すブロック図である。
【図２】通信システムの構成例を示すブロック図である。
【図３】最適サービスリスト決定機能部の動作を示すフローチャートである。
【図４】目標値最大化サービスリスト選出機能部の動作を示すフローチャートである。
【図５】サービスリスト算出装置の変形例を示すブロック図である。
【図６】第２の実施の形態のサービスリスト算出装置の構成を示すブロック図である。
【図７】網の１ボトルネックモデルを示す模式図である。
【図８】ネットワークワイドモデルを示す模式図である。
【図９】従来技術のサービスリスト算出装置の構成を示すブロック図である。
【符号の説明】
１０ 最適サービスリスト決定機能部
１１，１２，１３ 情報取得部
１４ 選択確率算出部
１５ 単位サービス網資源必要量算出部
１６ 論理パス目標関数値算出部
１７ 最適論理パス選択部
１８ 目標値最大化サービスリスト選出部
２１ 履歴データベース
２２ 計算条件データベース
２３ 需要予測機能部
２４ 選択確率モデル化機能部
５０ ユーザ端末
１００ ネットワーク
１０１ サービスリスト算出装置
１０２ サービスリスト提示システム
１０３ ユーザリクエスト受付システム
１０４ ＱｏＳ提供システム[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a service list selecting apparatus and method, a program, and a recording medium applied to a high-speed computer-to-computer communication network that provides a plurality of quality transfer services.
[0002]
[Prior art]
For example, in a communication network that realizes high-speed communication between computers, if a single class of quality assurance transfer service is to be provided to a specific user who has contracted for a fee, the service provider must provide the transfer service. Prices need to be set appropriately.
That is, it is highly probable that each user's decision whether to subscribe to the transfer service changes depending on the price of the transfer service. In addition, with the change in the number of contracted users, the amount of resources such as communication network hardware required for providing the service changes, and the income received by the business operator also changes significantly.
[0003]
In practice, since the amount of network resources is finite, the service provider must consider the change in service contract probability according to the price of the user within the range of available network resources, and It is assumed that the price is determined to maximize.
Non-Patent Document 1 and Non-Patent Document 2 are known as conventional techniques relating to such a method.
[0004]
Although Non-Patent Document 1 deals with a model in which the price changes according to the number of simultaneous connections as a charging model, it can be applied to a model in which the price does not change regardless of the number of simultaneous connections in the same period. .
When the method of Non-Patent Document 1 is implemented, a demand forecasting function unit, a user selection modeling unit, and an optimal price determination unit are provided, and the time is divided for each fixed period and a new period is set at the start of each period. Calculate the price to use.
[0005]
First, the demand forecasting function unit reads, for each segment s, the history of user calls actually made in the past period from the history database, and calculates the user occurrence rate λs in the new period by using a Kalman filter that inputs them. Then, λs is transmitted to the optimal price determination unit.
Next, in the user selection modeling unit, for each segment s, a past user service selection history and the price at that time are stored in a history database for a service selection probability formula that inputs a price including an unknown parameter prepared in advance. , And using the data as input, the maximum likelihood estimation method is used to calculate the parameters of the service selection probability equation so that the selection history is the best. As a result, the selection probability function Ps (price) of the service i having the price of the traffic of the segment s as the input is created, and λs is transmitted to the optimal price determination unit.
[0006]
Then, the optimal price determination unit receives the occurrence rate λs from the demand prediction function unit, receives the selection probability function Ps (I, price) from the user selection modeling unit, and determines them and the available network resource amount R and the single service. The maximum number of connections N is set to (N = R / r) using the network resource amount r required for providing, and the occurrence rate at which the user makes a contract and the network resource is required is (λ = Σsλs × Ps ( ), And the termination rate μ is treated as a single queue model of M / M / N (0) which is the most basic in traffic theory, assuming that the termination rate μ is given in advance regardless of the segment s. As a result, the income is represented by a price function, and the optimal price can be calculated.
[0007]
On the other hand, in Non-Patent Document 2, high-speed computer-to-computer communication that provides a plurality of transfer services with different transfer qualities at a differentiated price by using a technology such as Differentiated Services (hereinafter Diffserv) or Integrated Services (hereinafter Intserv). A method for setting an optimum service list in a network has been proposed.
Here, each provided transfer service is called a service class, and a set of price and quality of all service classes is called a service list.
[0008]
In this case, the network resources required to provide one service contract of each service class, that is, the processing capacity of the computer and the bandwidth of the communication line differ for each service class.
As described above, in a network that provides services of a plurality of qualities, the user judges which service class to use by comparing the contents of the service list presented by the network operator. For this reason, it is considered that the distribution of the selection probabilities of the service classes respectively selected as the transfer services actually used by the respective users changes according to the contents of the service list presented by the network operator.
[0009]
In other words, when the service list changes, the service selection probability changes, so the distribution of the number of service selection users also changes. If the number of service selection users changes, both the revenue of the network operator and the amount of network resources required for providing the service change.
For this reason, the network operator calculates or selects an optimal service list required to realize maximum income within the range of available network resources while considering the change in the user's selection probability for the service list. There is a need.
[0010]
The technology of Non-Patent Document 2 includes a demand prediction function unit, a user selection probability modeling function unit, and an optimum service list determination function unit, as shown in FIG.
As for demand, it is considered that there is a periodic fluctuation depending on the day and time. Therefore, the demand prediction function unit of Non-Patent Document 2 reads out the number of user calls for each segment s in the past one period from the history database and inputs them to the Lehman filter by the same method as the conventional technology. Thus, the number of calls in the new period is predicted, and the predicted occurrence rate λs is further calculated.
[0011]
In Non-Patent Document 2, only the occurrence rate λs is targeted for prediction, but the termination rate μs is also predicted by reading the number of finished calls from the history database and inputting them to the Kalman filter to predict the number of finished calls in the new period. In addition, it is also possible to calculate the prediction end rate μs.
Further, the user selection probability modeling function unit of Non-Patent Document 2 reads out the past service selection history for each segment s and the service list contents when each is selected from the history database, and uses them as inputs to perform maximum likelihood estimation. , A service selection probability function Ps (i, SL) relating to a service list for selecting a service i for each segment s can be created.
[0012]
Then, the optimal service list determination function unit of Non-Patent Document 2 calculates a predicted occurrence rate λs, a predicted end rate μs, a service selection probability function Ps (i, SL), a bottleneck resource amount R read from a database, and a certain service. Unit service network resource amount function r required for each user of each service class i to which list SL is input_iUsing the reference of (SL) and a set of candidates that can be taken as SL, a service list SL that becomes an optimal solution is determined.
[0013]
The technology of Non-Patent Document 2 assumes a one bottleneck link model as shown in FIG. 7 in which a network resource serving as a bottleneck exists only at one place.
[Non-patent document 1]
Noriaki Kamiyama, "Charging method that satisfies fairness for IP network resource allocation service", IEICE, Vol. J83-B, No. 7, pp. 999-1011, July 2000.
[Non-patent document 2]
Ikuo Yamazaki et al., "Study of Income Maximization Method in Multi-Service Class Network", IEICE Technical Report, CS2002-96 (Co-hosted NS2002-141, IN2002-85), pp71-76, October 2002.
[0014]
[Problems to be solved by the invention]
In the prior art as described above, all traffic always passes through a certain link in the network, and other than that point does not become a bottleneck of the network resource amount. Is going on.
[0015]
However, in an actual communication network, there is a two-dimensional spread as in the network wide model shown in FIG. 8, and traffic generally passes through various routes.
Further, in order to provide a plurality of transfer services with different transfer qualities, the amount of traffic generated from one network end node (communication device terminating the network) and terminated by another network end node is determined by which route Need to be managed and controlled.
[0016]
In such a case, since it is not scalable to perform bandwidth management in a fine unit such as a flow at each core node that relays communication, the traffic between network edge nodes is grasped as one bundle, and the logical path is determined. It is considered that it is preferable to extend the bandwidth between the network edge nodes, allocate the bandwidth to the logical path, and perform the allocated bandwidth management in the relay core node for each path.
[0017]
In this case, regarding traffic (call) admission control performed when a new call is generated, management such as whether the available bandwidth within the logical path setting bandwidth is equal to or more than the required bandwidth of the generated call is performed by the network. It is conceivable to perform distributed processing at the end node.
In the one-link bottleneck link model assumed in the prior art, since the network resource amount of the bottleneck link is constant, the only parameter that affects the revenue is the service list. However, in a network-wide model to be assumed in an actual communication network, available network resources, that is, logical paths to be allocated between the network end nodes under given conditions of network topology and available bandwidth of each link. The allocation status, that is, the allocation status of the route and the bandwidth of the logical path, will affect the revenue together with the service list.
[0018]
Therefore, in order for the network operator to maximize the revenue, it is necessary to optimize not only the service list but also the assignment state of each logical path at the same time.
An object of the present invention is to provide a service list selection device and method, a program, and a recording medium that enable optimization of a service list and optimization of an allocation state of each logical path.
[0019]
[Means for Solving the Problems]
According to a first aspect of the present invention, when there are a plurality of service classes having different communication service qualities, a plurality of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. A service list selecting apparatus used to select a desired service list SL from a plurality of service lists SL, wherein at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. In this case, the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL), a second information acquisition unit for acquiring a set of candidates that can be taken as the service list SL, and the selection probability formula P for the input service list SL.^sA selection probability formula calculation unit for calculating (j, SL); and a unit service network resource amount function r for an input service list SL._j(SL), a unit service network resource amount function calculating unit, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}And logical path candidate path^{(I, e)}Is selected as the target value Target (SL, path^{(I, e)}), And 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. The total revenue obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. And an optimal logical path selection unit for calculating an optimal target function value and an optimal logical path assignment in the case of performing the above, and initializing an optimal target function value for the first service list SL, and assigning each of the other service list SL candidates On the other hand, the logical path candidate path when the service list SL is fixed^{(I, e)}And calculating the optimum target function value and the optimum logical path by the optimum logical path selecting unit and updating the optimum target function value with the calculated values in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized, and a target value maximizing service list selecting unit for selecting an assignment state of a logical path are provided.
[0020]
According to the first aspect, when providing a plurality of quality communication services on an actual communication network corresponding to a network wide model, not only optimization of a service list but also optimization of an assignment state of each logical path at the same time. , Can maximize income.
The “band information indicating the available bandwidth of each link constituting the network” handled by the second information acquisition unit is, for example, the available bandwidth of the connection link existing between the node n1 and the node n2. It can be represented as a matrix R (n1, n2) whose width is each element.
[0021]
Further, when a service list is set, a finite logical path candidate constituted by a combination of discrete set bandwidths and routes is obtained, and when the service list is fixed, 0- In order to enable formulation as a one-integer programming problem, it is possible to derive an optimal solution. By performing this process for all service lists, an optimal service list that ultimately maximizes revenue and It is possible to calculate the optimum logical path assignment.
[0022]
According to a second aspect, when there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. A service list selecting apparatus used to select a desired service list SL from a plurality of service lists SL, wherein at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. In this case, the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL) reference, a set of candidates that can be taken as the service list SL, a second information acquisition unit that acquires cost information per unit band of each link, and the selection probability for the input service list SL Formula P^sA selection probability formula calculation unit for calculating (j, SL); and a unit service network resource amount function r for an input service list SL._j(SL), a unit service network resource amount function calculating unit, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}Is calculated, and using the cost information per unit band of each link, the cost for the set band in all the passing links is calculated as the expected expenditure, and the logical path candidate path is calculated.^{(I, e)}Is selected by subtracting the expected expenditure from the expected income obtained in each logical path when the target is selected as the target value Target (SL, path^{(I, e)}), And 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. , The total profit obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. An optimal logical path selector for calculating an optimal target function value and an optimal logical path assignment in the case of
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path when the service list SL is fixed.^{(I, e)}And calculating the optimum target function value and the optimum logical path by the optimum logical path selecting unit and updating the optimum target function value with the calculated values in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized, and a target value maximizing service list selecting unit for selecting an assignment state of a logical path are provided.
[0023]
According to claim 2, when providing a plurality of quality communication services on an actual communication network corresponding to a network wide model, not only optimization of a service list but also optimization of an assignment state of each logical path at the same time, In addition, maximization of profit can be realized by reflecting the expected expenditure in the calculation.
A third aspect of the present invention provides the service list selection apparatus according to the first or second aspect, wherein a history database that retains a history of past call occurrences and call durations and a history of service selection regarding the pair (i, e) of the network edge nodes. And the predicted call occurrence rate λ based on the history information read from the history database.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability equation P representing the probability that the segment s selects the service class j from the service list SL based on the history information read from the history database.^sA user selection modeling unit for predicting (j, SL) is further provided.
[0024]
According to claim 3, based on the history information held in the history database, the predicted call occurrence rate λ^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And the selection probability formula P^sSince (j, SL) is predicted and obtained by calculation, it is possible to perform optimization by reflecting, for example, a change in a use situation or a selection situation in an actual communication network.
Further, the predicted call occurrence rate λ for each pair (i, e) between edge nodes^{s (i, e)}And predicted call termination rate μ^{s (i, e)}Is used, a traffic model of a call requesting each service class can be predicted for each pair (i, e).
[0025]
According to a fourth aspect, when there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. A service list selection method used to select a desired service list SL from a plurality of service lists SL, wherein at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. In this case, the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing a usable bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL), a second information acquisition procedure for acquiring a set of candidates that can be taken as a service list SL, and the selection probability formula P for the input service list SL.^sA selection probability formula calculation procedure for calculating (j, SL); and a unit service network resource amount function r for the input service list SL._j(SL), a unit service network resource amount function calculating procedure, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}And logical path candidate path^{(I, e)}Is selected as the target value Target (SL, path^{(I, e)}), And a 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. The total revenue obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. The optimal logical path selection procedure for calculating the optimal target function value and the optimal logical path assignment in the case of performing the above, and initializing the optimal target function value for the first service list SL, and On the other hand, the logical path candidate path when the service list SL is fixed^{(I, e)}Is used to calculate the optimal target function value and the optimal logical path in the optimal logical path selection procedure, and to update the optimal target function value with the calculated value in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path are provided.
[0026]
According to claim 4, the same result as that of the device of claim 1 is obtained.
According to a fifth aspect, when there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. A service list selection method used to select a desired service list SL from a plurality of service lists SL, wherein at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. In this case, the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing a usable bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL) reference, a set of candidates that can be taken as a service list SL, a second information acquisition procedure for acquiring cost information per unit band of each link, and the selection probability for the input service list SL Formula P^sA selection probability formula calculation procedure for calculating (j, SL); and a unit service network resource amount function r for the input service list SL._j(SL), a unit service network resource amount function calculating procedure, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}Is calculated, and using the cost information per unit band of each link, the cost for the set band in all the passing links is calculated as the expected expenditure, and the logical path candidate path is calculated.^{(I, e)}Is selected by subtracting the expected expenditure from the expected income obtained in each logical path when the target is selected as the target value Target (SL, path^{(I, e)}), And a 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. , The total profit obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. The optimal logical path selection procedure for calculating the optimal target function value and the optimal logical path assignment in the case of performing the above, and initializing the optimal target function value for the first service list SL, and On the other hand, the logical path candidate path when the service list SL is fixed^{(I, e)}And calculating the optimal target function value and the optimal logical path by the optimal logical path selection procedure, and updating the optimal target function value with the calculated values in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path are provided.
[0027]
According to claim 5, the same result as the device of claim 2 is obtained.
According to a sixth aspect of the present invention, in the service list selection method of the fourth or fifth aspect, a history of past call occurrence and call duration and a history of service selection regarding the pair (i, e) of the network edge nodes are stored in advance. The predicted call occurrence rate λ based on the history information read from the history database.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL based on the demand forecasting procedure for calculating the service class SL and the history information read from the history database.^sA user selection modeling procedure for estimating (j, SL).
[0028]
According to claim 6, the same result as that of the device of claim 3 is obtained.
According to a seventh aspect, when there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. In a computer-executable program for selecting a desired service list SL from a plurality of service lists SL, at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. Is the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing a usable bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL), a second information acquisition procedure for acquiring a set of candidates that can be taken as a service list SL, and the selection probability formula P for the input service list SL.^sA selection probability formula calculation procedure for calculating (j, SL); and a unit service network resource amount function r for the input service list SL._j(SL), a unit service network resource amount function calculating procedure, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}And logical path candidate path^{(I, e)}Is selected as the target value Target (SL, path^{(I, e)}), And a 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. The total revenue obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. The optimal logical path selection procedure for calculating the optimal target function value and the optimal logical path assignment in the case of performing the above, and initializing the optimal target function value for the first service list SL, and On the other hand, the logical path candidate path when the service list SL is fixed^{(I, e)}Is used to calculate the optimal target function value and the optimal logical path in the optimal logical path selection procedure, and to update the optimal target function value with the calculated value in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path are provided.
[0029]
By executing the program according to claim 7 using a predetermined computer, the same result as that of the apparatus according to claim 1 can be obtained.
In the case where there are a plurality of service classes having mutually different communication service qualities, a plurality of service lists SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance. In a computer-executable program for selecting a desired service list SL from a plurality of service lists SL, at least one of a section related to a user, a section related to an application to be used, and a section related to content is represented as a segment s. Is the predicted call occurrence rate λ for a pair (i, e) of one network edge node i and another network edge node e.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL.^s(J, SL), a network topology representing a network configuration, bandwidth information representing a usable bandwidth of each link constituting the network, and a logical path of each logical path on the network. Constraint conditions on bandwidth candidates and route candidates, and a unit service network resource amount function r representing a network resource amount required for each user of each service class j for a certain service list SL_j(SL) reference, a set of candidates that can be taken as a service list SL, a second information acquisition procedure for acquiring cost information per unit band of each link, and the selection probability for the input service list SL Formula P^sA selection probability formula calculation procedure for calculating (j, SL); and a unit service network resource amount function r for the input service list SL._j(SL), a unit service network resource amount function calculating procedure, and an input service list SL.^{s (i, e)}And the predicted call termination rate μ^{s (i, e)}And the selection probability formula P^s(J, SL) calculation result and unit service network resource amount function r_jUsing the calculation result of (SL) and a finite number of discrete value setting band candidates and a finite number of route candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path, Logical path candidate path^{(I, e)}Is calculated, and using the cost information per unit band of each link, the cost for the set band in all the passing links is calculated as the expected expenditure, and the logical path candidate path is calculated.^{(I, e)}Is selected by subtracting the expected expenditure from the expected income obtained in each logical path when the target is selected as the target value Target (SL, path^{(I, e)}), And a 0-1 variable indicating whether or not each route is selected for the input service list SL for all logical path candidates between all network end nodes. Prepared and the calculated target value Target (SL, path) of each logical path candidate^{(I, e)}), The constraint that the total of the selected logical path setting bands passing through each link does not exceed the available link band, and the constraint that one of the logical path candidates is selected. , The total profit obtained from each logical path is defined as a target function value TargetValue, and the maximization thereof is formulated as a 0-1 integer programming problem, and the service list SL is fixed by solving this problem. The optimal logical path selection procedure for calculating the optimal target function value and the optimal logical path assignment in the case of performing the above, and initializing the optimal target function value for the first service list SL, and On the other hand, the logical path candidate path when the service list SL is fixed^{(I, e)}And calculating the optimal target function value and the optimal logical path by the optimal logical path selection procedure, and updating the optimal target function value with the calculated values in order for all the service list SL candidates. The present invention is characterized in that a service list SL in which the target function value TargetValue is finally maximized and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path are provided.
[0030]
By executing the program of claim 8 using a predetermined computer, the same result as that of the apparatus of claim 2 can be obtained.
According to a ninth aspect of the present invention, in the program according to the seventh or the eighth aspect, a history database in which a history of past call occurrence and call duration and a history of service selection regarding the pair of network edge nodes (i, e) are stored in advance. Based on the history information read from the^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And a selection probability formula P representing the probability that the segment s selects the service class j from the service list SL based on the demand forecasting procedure for calculating the service class SL and the history information read from the history database.^sA user selection modeling procedure for estimating (j, SL).
[0031]
By executing the program of claim 9 using a predetermined computer, the same result as that of the apparatus of claim 3 can be obtained.
According to a tenth aspect of the present invention, there is provided a computer-readable recording medium storing the program according to the seventh, eighth, and ninth aspects.
[0032]
By executing the program read from the recording medium of the tenth aspect on a predetermined computer, the same result as that of the apparatus of the first to third aspects can be obtained.
[0033]
BEST MODE FOR CARRYING OUT THE INVENTION
(First Embodiment)
One embodiment of a service list selection device and method, a program and a recording medium according to the present invention will be described with reference to FIGS. This form corresponds to claims 1, claim 3, claim 4, claim 6, claim 7, claim 9, and claim 10.
[0034]
FIG. 1 is a block diagram showing the configuration of the service list calculation device of this embodiment. FIG. 2 is a block diagram illustrating a configuration example of a communication system. FIG. 3 is a flowchart showing the operation of the optimum service list determination function unit. FIG. 4 is a flowchart showing the operation of the target value maximizing service list selecting function unit. FIG. 5 is a block diagram showing a modification of the service list calculation device.
[0035]
In this embodiment, the first information acquisition unit, the second information acquisition unit, the selection probability formula calculation unit, the unit service network resource amount function calculation unit, the logical path target function value calculation unit, and the optimum logical path selection unit according to claim 1 The target value maximizing service list selecting unit includes an information obtaining unit 11 (12), an information obtaining unit 13, a selection probability calculating unit 14, a unit service network resource required amount calculating unit 15, a logical path target function value calculating unit 16, This corresponds to the optimum logical path selecting unit 17 and the target value maximizing service list selecting unit 18.
[0036]
The history database, the demand forecasting unit, and the user selection modeling unit in claim 3 correspond to the history database 21, the demand forecasting function unit 23, and the selection probability modeling function unit 24, respectively.
[0037]
In this embodiment, it is assumed that the present invention is applied to a communication system as shown in FIG. That is, the service list calculation device 101, the service list presentation system 102, the QoS providing system 104, the user request receiving system 103, and the user terminal 50 are connected to each other via the network 100.
The service list calculation device 101 implements the present invention, and selects one service list SL considered to be optimal from a plurality of predetermined service list SL candidates. One service list SL is configured as follows, for example.
[0038]
class. 1: 1Mbps, 10 yen / 5 minutes
class. 2: 2Mbps, 15 yen / 5 minutes
class. 3: 3Mbps, 20 yen / 5 minutes
In this example, the service list SL has three service classes class. 1, class. 2, class. 3, and different service qualities and prices are assigned to the respective service classes.
[0039]
In this example, the service list calculation device 101 repeatedly selects (calculates) the service list SL at regular intervals determined as a control period. The service list SL selected by the service list calculation device 101 is presented to the user. Further, when the service list SL changes, the latest contents of the service list SL are reflected on the communication quality and the charging state of the actual service.
[0040]
The timing at which the service list SL selected by the service list calculating apparatus 101 is reflected on the content of the presentation to the user and the content of the service may be performed in each cycle of the control period, or may be performed in each specific period longer than that. (For example, every month).
The service list presentation system 102 presents one service list SL selected by the service list calculation device 101 to the user. The QoS (quality) providing system 104 realizes the communication quality of each of a plurality of service classes according to the contents of the service list SL selected by the service list calculation device 101.
[0041]
The user request reception system 103 receives a call from a user and a call termination. Further, the user request reception system 103 executes reception and rejection of a request from the user regarding service provision at the time of calling and terminating the call, and transmits the contents to the QoS providing system 104.
Further, the user request receiving system 103 identifies the user who issued the request, further identifies the segment s to which the user belongs, and records the following information as a history in the history database (21) in the service list calculation device 101. The segment s represents at least one of a section related to a user, a section related to an application to be used, and a section related to content.
[0042]
Outgoing user ID
Segment ID
Control period ID
Calling time
End time
Duration
Selected service class
Originating edge node ID
Destination edge node ID
Therefore, by examining the contents of the history database by designating the originating edge node ID, the terminating edge node ID, the originating segment ID, and the past control period ID, the user belonging to the designated segment can be identified by the designated edge. It is possible to obtain information such as the number of calls that have occurred within the control period specified between nodes and the average duration of calls that have ended.
[0043]
The QoS providing system 104 allocates each logical path to the network device. As a logical path to be controlled, a VP (Virtual Path) in an ATM network or an LSP (Label Switched Path) in an MPLS (Multiple Protocol Label Switching) network is assumed. In the case of an MPLS network, it is possible to explicitly assign an LPS.
[0044]
The service list calculation device 101 includes a history database 21, a demand prediction function unit 23, a selection probability modeling function unit 24, an optimal service list determination function unit 10, and a calculation condition database 22, as shown in FIG. The demand prediction function unit 23 includes an occurrence rate prediction unit and an end rate prediction unit.
The optimum service list determination function unit 10 includes information acquisition units 11, 12, 13, a selection probability calculation unit 14, a unit service network resource required amount calculation unit 15, a logical path target function value calculation unit 16, and an optimum logical path selection unit. 17 and a target value maximizing service list selecting unit 18.
[0045]
The components of each block shown in FIG. 1 can be actually realized as hardware or can be realized as a program executed by a computer.
First, the basic operation of the service list calculation device 101 will be described.
The above-mentioned history information is sequentially written and accumulated in the history database 21 by the user request reception system 103. Further, the contents of the service list SL selected by the service list calculation device 101 at each time are also stored in the history database 21.
[0046]
The calculation condition database 22 holds information on a plurality of service lists SL created in advance as candidates. Further, a unit service network resource amount function r required per user for the j-th service class of a certain service list SL_j(SL), logical path candidate conditions, network topology, and usable link bandwidth information are also determined in advance and held in the calculation condition database 22.
[0047]
The selection probability modeling function unit 24 reads the past service selection history of the user and the contents of the service list SL at the time of selection from the history database 21 for each segment s, and uses the maximum likelihood estimation method based on the information. A function P representing the probability that the user selects the service class j of the service list SL for each segment s^s(J, SL), that is, a service selection probability formula is created.
[0048]
It is considered that the actual demand for the service periodically fluctuates greatly depending on the day of the week and the time. Therefore, the occurrence rate prediction unit in the demand prediction function unit 23 determines from the history database 21 the call occurrence count history y regarding the traffic belonging to each segment s during the period (n-1) that has just ended.^s _n-1And input them to the Kalman filter to obtain the predicted number of call occurrences y in the next period n.^{s ＾} _nIs calculated. Further, the occurrence rate prediction unit calculates a predicted call occurrence rate (λ^s= Y^{s ＾} _n/ ΔT) (ΔT is the length of the period)^sIs calculated and transmitted to the optimum service list determination function unit 10. Y by force Le Mans filter^{s ＾} _nThe calculation method of will be described later.
[0049]
Further, the termination rate prediction unit in the demand prediction function unit 23 calculates, from the history database 21, the call average duration history h for the user belonging to each segment s during the period (n−1) that has just ended.^s _n-1, And inputting the information to the Kalman filter, the predicted call average duration h in the next period^{s ＾} _nIs calculated.
Then, the termination rate prediction unit calculates the prediction termination rate μ^sTo (μ^s= 1 / h^{s ＾} _n) And transmit it to the optimum service list determination function unit 10.
[0050]
Predicted end rate μ^s, It is possible to more accurately predict the traffic model for each segment, and it is also possible to automatically follow a change in the traffic model of the user. This predicted end rate μ^sAlso uses the optimum service list determination function unit 10.
Y by Kalman filter^{s ＾} _nThe method of calculating is described below.
[0051]
y^{s ＾} _nIs calculated by the following equation (1-1).
y^s＾_n= X_n ^tα_n-1            ... (1-1)
x_nIs called a dummy variable vector.
[0052]
Assuming that the number of control cycles is m and section n belongs to the k-th (0 ≦ k ≦ m−1) section in the control cycle, x_nIs an m-dimensional vector and is represented by the following equation.
x_n= (1, δ_1k, ..., δ_ik, ..., δ_{(M-1) k}) (1-2)
However,
δ_ik: 1 only when i = k, 0 otherwise
t: transpose of matrix and vector
Here, the dummy variable vector x_nIs multiplied by a weight vector α, and the characteristic of the Kalman filter is that α is sequentially updated. It is realized by the following update formula.
(Equation 1)

α_n: Estimated value of weight vector α in control section n
P_n: Α_nCovariance matrix (m × m dimensions)
K_n: Kalman gain (m-dimensional vector)
Q_n: Covariance matrix of system noise (m × m dimensions)
In practice, Q_nAs a positive definite matrix independent of n (eg, Q_n= 0.05I), and α is used as an initial value.₀Is the zero vector, (P₀= 1000I).
[0053]
In this example, at the end of the control section n, the demand forecasting function unit 23^s _nIs read out from the history database 21 and α is calculated using Expressions (1-3) to (1-6)._nTo update. Further, when the control section n is incremented and a new control section n is started, y is calculated according to the equation (1-1).^{s ＾} _nIs calculated.
If the control is concentrated at the start of the control section, the following processing will be performed when the control section (n-1) ends and the control section n starts. First, the observed number of call occurrences y^s _n-1Is read from the history database 21 and α is calculated according to the following equations (2-3) to (2-6)._n-1To update. And the updated α_n-1And using equation (1-1), y^{s ＾} _nIs calculated.
(Equation 2)

Average call duration h^{s ＾} _nFor the prediction of^s _nH^s _nBy replacing α with α_n-1And the updated α_n-1Using h^{s ＾} _nCan be realized by calculating
The demand prediction function unit 23 performs demand prediction on a pair (i, e) of one network edge node i and another network edge node e between the nodes.
[0054]
The optimal service list determination function unit 10 shown in FIG. 1 performs the processing shown in FIG.
In step S01 of FIG. 3, the information acquisition unit 11 of the optimum service list determination function unit 10 sends the predicted call occurrence rate λ for each network edge node pair (i, e) from the demand prediction function unit 23.^{s (i, e)}And predicted call termination rate μ^{s (i, e)}To get.
In step S02, the information acquisition unit 12 of the optimal service list determination function unit 10 sends a selection probability expression P representing the probability that the segment s selects the service class j from the service list SL from the selection probability modeling function unit 24.^s(J, SL) is obtained.
[0055]
In step S03, the information acquisition unit 13 of the optimal service list determination function unit 10 acquires the following information from the calculation condition database 22.
(1) Network topology
(2) Available bandwidth of each link (for example, a matrix R (n1, n2) in which the available bandwidth of a connection link existing between the nodes n1 and n2 is an element)
Constraints on bandwidth candidates and route candidates for each logical path
(3) A unit service network resource amount function r representing a network resource amount required for one user of each service class j of a certain service list SL to be input._j(SL) Reference
(4) Set of candidates that can be taken as SL
Steps S01, S02, and S03 may be sequentially performed in any order, or may be performed simultaneously.
[0056]
In step S04, based on the information obtained in the previous steps S01 to S03, the target value maximizing service list selecting unit 18 selects the service list having the maximum target value from all the SL candidates and the optimal logical path at that time. Select the assignment status.
[0057]
Actually, step S04 is realized by the processing of the optimum service list determination function unit 10. An outline of the operation of the optimum service list determination function unit 10 is as shown in FIG. Hereinafter, the operation of FIG. 4 will be described.
In step S10, a variable Max for storing the maximum target value is initialized to zero.
[0058]
For each of the service list SL candidates acquired by the information acquisition unit 13 from the calculation condition database 22, the processing after step S11 is sequentially performed. In step S11, it is checked whether there are any unprocessed service list SL candidates.
If there is an unprocessed service list SL candidate, the process proceeds from step S11 to S12. If not, the process ends.
[0059]
In step S12, one service list SL candidate is extracted from the unprocessed candidate group, and the number of service classes included in the service list SL, that is, the class number J is recognized.
In the next step S13, the unit service network resource required amount calculating unit 15 is used to calculate the unit service network resources for each service class j for all the elements (service classes) included in the selected service list SL candidates. Required amount r_j(SL) is calculated.
[0060]
In step S14, the selection probability calculation unit 14 is used to calculate a selection probability expression Ps (j, SL) for all segments s and all service classes j.
Note that either one of steps S13 and S14 may be performed first, or may be performed simultaneously.
In the next step S15, using the logical path target function value calculation unit 16, the predicted call occurrence rate λ obtained from the demand prediction function unit 23 for the logical path allocated between each network edge node pair (i, e).^{s (i, e)}And predicted call termination rate μ^{s (i, e)}And the selection probability formula P calculated by the selection probability calculation unit 14^s(J, SL) and the required unit service network resource amount r calculated by the required unit service network resource amount calculation unit 15_j(SL) based on a combination of a finite number of discrete value setting bandwidth candidates that meet the constraints on the bandwidth candidates and route candidates of each logical path acquired from the calculation condition database 22, and a finite number of route candidates Finite logical path candidate set Path^{(I, e)}And set the Path^{(I, e)}Is calculated as the target value Target (SL, (i, e, r, b)) obtained for each logical path that would be expected if it were selected.
[0061]
Here, r represents the ID (identifier) of the route candidate, and b represents the ID in the route candidate. That is, the logical path candidate set Path is expressed by (i, e, r, b).^{(I, e)}Is specified.
In the next step S16, the optimal logical path selecting unit 17 is used to change the 0-1 variable, which is "1" when each path is selected and "0" when not selected, to all logical paths between all nodes. Each of the logical path target function values calculated by the logical path target function value calculating unit 16 and the target value Target (SL, (i, e, r, b)) of each logical path, is used to select each logical path. Under the constraint that the total of the logical path setting bandwidths does not exceed the available link bandwidth, and the constraint that only one of the logical path candidates is selected, the total revenue obtained from each logical path Is defined as a target function value TargetValue, and the maximization thereof is formulated as an 0-1 integer programming problem with an optimization condition. At that time, the target function value TargetValue is initialized to Max.
[0062]
In step S17, the 0-1 integer programming problem formulated in step S16 is solved so as to be optimized, and a target function value TargetValue is obtained.
In step S18, the resulting target function value TargetValue is compared with Max. The process proceeds to the next step S19 only when the condition of (TargetValue> Max) is satisfied.
[0063]
In step S19, Max is updated to TargetValue. Further, the service list SL and the respective logical path allocation states (paths and bandwidths) at that time are stored as the optimum service list SL and the optimum logical path allocation, respectively.
When the processing after step S11 is sequentially executed for all the service list SL candidates, as a result, the optimum service list SL and the optimum logical path assignment are finally calculated.
[0064]
The optimum logical path selecting unit 17 formulates the 0-1 integer programming problem as follows.
First, the following parameters are defined for the given physical network. C (n1, n2): If the link (n1, n2) exists, the available bandwidth is allocated, and if not, 0 is allocated. The link (n1, n2) means a link directly connecting the node n1 and the node n2. The link set is (LINK).
[0065]
All: All sets of logical path candidates. The ID of each logical path is (i, e, r, b), and among the logical path candidates set between the network edge node i and the network edge node e, (route candidate ID = r, bandwidth candidate ID = B).
numR_{(I, e)}: The number of route candidates for the logical path (i, e), and the ID is r.
numB_{(I, e)}: The number of bandwidth candidates for the logical path (i, e), and the ID is b.
[0066]
g^{(I, e) r} _{(N1, n2)}: 1 if the path candidate ID = r of the logical path (i, e) passes through the link (n1, n2), and 0 otherwise.
BW^{(I, e) b}: Assigned bandwidth of logical path (i, e) bandwidth candidate ID = b.
income_{(I, e)}(R, b, SL): When the service list is set to SL and the logical path (i, e, r, b) is set, the expected income value obtained from this logical path is obtained.
[0067]
target_{(I, e)}(R, b, SL): A target function value obtained from the logical path when the service list is defined as SL and the logical path (i, e, r, b) is determined. If your goal is income (income_{(I, e)}(R, b, SL)).
x^{(R, b)} _{(I, e)}: 0-1 which is "1" when the logical path (i, e) takes (candidate ID = r) as the route and (candidate ID = b) as the allocated bandwidth, and "0" otherwise. variable. This is the solution to be found.
[0068]
The parameters required for formulation are as described above. Since the optimization function and the goal are to maximize the sum of the expected revenues obtained in the e2e logical path, the following equation is obtained.
(Equation 3)

The expression (3-2) expresses an exclusion condition that only one of the LSP candidates taking into account both the path and the allocated bandwidth for the logical path (i, e) can be selected. ing. Equation (3-3) indicates that when each e2e logical path selects its own logical path candidate, the total bandwidth required by each logical path in the (n1, n2) link is smaller than the available link bandwidth. Represents a constraint that means small.
[0069]
In the formulas (3-1) to (3-4), g for all (i, e), all r, and all b^{(I, e) r} _{(N1, n2)}And BW^{(I, e) b}And the target in that case_{(I, e)}Once (r, b, SL) is determined, the variable is x^{(R, b)} _{(I, e)}Only the 0-1 integer programming problem. This can be solved by an existing solution such as a branch and bound method.
In the service list calculation device of this embodiment, the following is assumed as a specific example of the constraint condition regarding the bandwidth candidate and the route candidate of each logical path read by the logical path target function value calculation unit 16.
[0070]
Examples of logical path route candidate conditions:
[Route Candidate Condition (1)] A route whose hop count is equal to or less than “the shortest route hop count + α (given constant positive integer)”.
[Path Candidate Condition (2)] A path whose total link delay is equal to or less than “minimum total link delay + β (given constant positive integer)”.
[0071]
Examples of logical path bandwidth candidate conditions:
[Bandwidth candidate condition (1)] A finite number of fixed bandwidths prepared in advance for each logical path or for all logical paths in advance are set as candidates.
[0072]
[Bandwidth Candidate Condition (2)] The minimum unit that can be set for a logical path is determined as BwUnit, and for each of a limited number of target blocking rates prepared in advance for each logical path or for all logical paths, The maximum bandwidth at which the expected call loss rate is equal to or less than the target call loss rate is set as a bandwidth candidate. Each time the service list changes, the maximum bandwidth that satisfies each target blocking rate is dynamically calculated.
[0073]
[Bandwidth Candidate Condition (3)] The settable unit of the logical path is determined as BwStep, and the expected loss probability is calculated for one target loss probability prepared in advance for each logical path or common to all the logical paths. The maximum bandwidth that is equal to or less than the target blocking rate is set as the reference bandwidth, and the bandwidth is determined as a bandwidth candidate including a predetermined fixed number of discrete bandwidths that can be set above and below the reference bandwidth. Each time the service list changes, a reference bandwidth and its upper limit discrete bandwidth are dynamically calculated.
[0074]
Here, when the [path candidate condition (1)] and the [bandwidth candidate condition (1)] are applied, it is assumed that a call occurs at random and the duration follows an exponential time. (Target (SL, path) in the logical path target function value calculation unit 16 when the list SL is assigned.^{(I, e)})) Is shown below.
First, a certain (i, e) is fixed. The logical path target function value calculation unit 16 extracts all routes that meet the [path candidate condition (1)]. From the node i, increase the number of hops one by one in order to create a tree-type connection diagram that does not loop, calculate the shortest path, and continue until the hop number reaches [shortest path + α] or reaches node e. Then, when there are no longer any branches to be extended, only the branches that have reached the node e are traced back to complete the path candidates. [Bandwidth candidate condition (1)] represents each of the read bandwidths as a bandwidth BW.^{(I, e) b}Just set it to
[0075]
Next, the calculation of (Target (i, e, r, b, SL)) will be described. When the income is set to the target value (Target (i, e, r, b, SL)), there is no difference between the routes, so the set bandwidth BW^{(I, e) b}Can be regarded as one bottleneck model.
[0076]
A case where a multi-call source traffic model is used will be described.
The selection probability formula P calculated by the selection probability calculation unit 14^sUsing (j, SL), the generated call volume a for service class j generated during (i, e) in the case of SL_j ^{(I, e)}Is calculated from the following equation.
a_j ^{(I, e)}= Σ_s｛Λ^{s (i, e)}× P^s(J, SL) / μ^{s (i, e)}・ ・ ・ (4-1)
Next, P^s(J, SL) and a_j ^{(I, e)}And BW^{(I, e) b}And r calculated by the unit service network resource required amount calculation unit 15_j(SL) and (0 ≦ Σ_j(N_j× r_j(SL)) ≦ BW^{(I, e) b}) Are calculated by the following equation: p (n1,..., Nj,..., NJ).
[0077]
p (n1,..., nj,..., nJ) = Π^J _j((A_j)^nj/ N_j! ) × p (0,0 ,,, 0) (4-2)
However,
p (0,... 0) = 1 / [Σ^NJ _{nj = 0}... Σ^N2 _{n2 = 0}Σ^N1 _{n1 = 0}Π^J _j((A_j)^nj/ N_j! )],
N_j= INT [BW^{(I, e) b}−Σ^J _{k = j + 1}(N_k× r_k(SL))) / r_j(SL))],
INT [a] represents integer conversion in which a is rounded down to the decimal point of a.
[0078]
Then, this logical path candidate path^{(I, e)}From the following formula.

However,
N_j= INT [BW^{(I, e) b}−Σ^J _{k = j + 1}(N_k× r_k(SL)) / r_j(SL)],
price_jIs the price per unit time of the class j included in the service list SL.
[0079]
It is also possible to model as a single call source model. First, the generated call volume a of each segment s during (i, e)^sIs calculated from the following equation.
a^{s (i, e)}= Λ^{s (i, e)}/ Μ^{s (i, e)}, A^{(I, e)}= Σ_sa^{s (i, e)}        ... (4-4)
The maximum number of states N is calculated by the following equation.

The meaning of this expression will be described. Now, assume that the number of call states for the total of all segments during (i, e) is n. Here, the number of calls from the segment s is more average ((a^s/ A) × n), and the number of calls for which the call of the segment s selects the service class j is ((a)^s/ A) × P^s(J, SL) × n).
[0080]
The call to select service class j is network resource r_j(SL), the total (Σ) for all service classes in all segments_j(Σ_s[A^s/ A × P^s(J, SL)] × r_jIt is considered that (SL)) × n) network resources are required. This is BW^{(I, e) b}If it is below, it can be accommodated, so the following relational expression holds.
Σ_j(Σ_s[A^s/ A × P^s(J, SL)] × r_j(SL)) × n ≦ BW^{(I, e) b}    ... (4-6)
Therefore, the maximum number of states N that can be taken as n can be estimated as follows.
[0081]
N = INT [BW^{(I, e) b}/ Σ_j(Σ_s[A^s/ A × P^s(J, SL)] × r_j(SL))] ... (4-7)
Then, the steady state probability p (n) for the state n (1 ≦ n ≦ N) is expressed by the following equation.
p (n) = (Aⁿ/ N! ) × p (0)
p (0) = 1 / [1 + Σ^N _n(Aⁿ/ N! )]
That is, it can be calculated by the steady state probability formula in the single queue model M / M / N (0). Since the objective function is the expected income obtained from (i, e, r, b), it is obtained from the following equation.
[0082]

This equation will be described.
As in the case described above, assume that the number of call states for all segments is n. At this time, the number of calls from the segment s can be estimated on average from the ratio of the generated call volume to ((as / A) × n), and the call of the segment s selects the service class j. The number is ((a^s/ A) × P^s(J, SL) × n).
[0083]
Thus, the total number of calls that select service class j for all segments s is (Σ_s｛N × (a^s/ A) × P^s(J, SL)｝), the revenue from all calls in state n is (Σ_j[Price_j× (Σ_s[N × (a^s/ A) × P^s(J, SL)])]).
Since this is multiplied by the state probability p (n) and added for all states, this means the expected value of the network revenue.
[0084]
Compare the calculation of target (i, e, r, b, SL) with the multi-call queuing model and the calculation with the single queuing model. In the former, the calculation accuracy is high in any situation, but the amount of calculation becomes enormous. The latter requires a small amount of calculation, but generally greatly degrades the calculation accuracy. However, in the latter case, where the call loss hardly occurs, the calculation accuracy is as high as that of the multi-call source model, so the bandwidth that can be used by the traffic subject to this control is the bandwidth required as a result of service selection. It is considered appropriate to use it in an overwhelming situation.
[0085]
As described above, when providing a plurality of service classes in a communication network that is a network-wide model, the demand between each network end node, the change in the service selection probability of the user based on the service list, and the logical path allocation state are taken into account. Thus, it is possible to determine a service list that maximizes network revenue and a logical path assignment state at that time.
[0086]
In the service list calculating device shown in FIG. 1, the call occurrence rate λ for each network edge node pair (i, e)^{s (i, e)}And call termination rate μ^{s (i, e)}And the selection probability formula P^sAlthough (j, SL) is predicted from the history information by the demand prediction function unit 23 and the selection probability modeling function unit 24, the call occurrence rate λ previously calculated or the like is calculated.^{s (i, e)}And call termination rate μ^{s (i, e)}And the selection probability formula P^sIf (j, SL) is held in the database as a constant, the configuration can be simplified as shown in FIG. 5 by omitting the demand prediction function unit 23 and the selection probability modeling function unit 24.
[0087]
(Second embodiment)
Another embodiment of the service list selection device and method, program and recording medium of the present invention will be described with reference to FIG. This embodiment corresponds to claims 2, 5, and 8.
FIG. 6 is a block diagram showing the configuration of the service list calculation device of this embodiment. This embodiment is a modification of the first embodiment. 6, elements corresponding to those in FIG. 1 are denoted by the same reference numerals. For the same parts as those in the first embodiment, the following description is omitted.
[0088]
As shown in FIG. 6, the service list calculation device of this embodiment also includes a history database 21, a calculation condition database 22B, a demand prediction function unit 23, a selection probability modeling function unit 24, and an optimum service list determination function unit 10.
As information held in the calculation condition database 22B, cost information for each link of the network (representing the cost per unit band of each link) is added. The optimal service list determining function unit 10 of this embodiment performs the calculation using the cost information for each link.
[0089]
The outline of the operation of the optimum service list determining function unit 10 is the same as that of FIG. 3, but in step S03, cost information for each link is also acquired from the calculation condition database 22B.
Further, the content of “selection of a target value maximizing service list” in step S04 is the same as that in FIG.
[0090]
Here, as in the case of the first embodiment, when the [path candidate condition (1)] and the [bandwidth candidate condition (1)] are applied, the occurrence of a call is random, and the duration is Is calculated according to the exponential time, and when a certain service list SL is assigned, the calculation formula (Target (i, e, r, b, SL)) relating to the profit in the logical path target function value calculation unit 16 is obtained. Is shown.
[0091]
Here, the following are defined as parameters.
Cost (n1, n2): Cost per unit time / band of link (n1, n2).
Income_{(I, e)}(R, b, SL): Expected revenue value obtained from the logical path when the service list is set to SL and the logical path (i, e, r, b) is selected.
Expenditure (i, e) (r, b, SL): When the service list is set to SL and the logical path (i, e, r, b) is selected, the set bandwidth of this logical path and its path Such expected cost values.
[0092]
In this case, the expected cost value Expenditure (i, e) (r, b, SL) is obtained by the following equation.

Also, the expected income value Income_{(I, e)}Since (r, b, SL) is as described above, the revenue (target (i, e) (r, b, SL)) is expressed by the following equation when the multi-call source traffic model is used.
[0093]

However,
N_j= INT [BW^{(I, e) b}−Σ^J _{k = j + 1}(N_k× r_k(SL)) / r_j(SL)],
price_jIs the price per unit time of the class j included in the service list SL.
[0094]
When the single call source queuing model is applied, the profit (target (i, e) (r, b, SL)) is expressed by the following equation.

As described above, when a plurality of service classes are provided in a communication network having a network wide model, a change in a user's service selection probability based on a demand between service nodes and a service list and a logical path assignment state are taken into consideration. Thus, it is possible to determine a service list that maximizes network profits and a logical path assignment state at that time.
[0095]
Note that the optimal service list and the optimal logical path assignment state calculated by the service list selection method of the present invention can be immediately applied to the next control period, but this is not always desirable.
For example, from the viewpoint of the user, the fact that the service list (charge system) for the next control period is undecided may be uneasy as a service. Therefore, in an actual communication service, it is conceivable to implement a service in which a service list to be applied is announced in advance for a certain period (for example, one week or one month).
[0096]
In such a case, it is conceivable that the system implementing the present invention is operated in the background as a service list proposal system, and uses the information of the calculated service list as a criterion for the network service provider. .
Also, regarding the logical path allocation control, it is possible to control the optimum logical path allocation state calculated by the method of the present invention to be fixed during the control period. It is technically possible to perform the reallocation control (increase or decrease of the band).
[0097]
Such an actual implementation situation may be appropriately changed according to the environment and necessity.
[0098]
【The invention's effect】
As described above, according to the present invention, when a plurality of service classes are provided in a communication network corresponding to the network wide model, the demand between each network end node and the change in the service selection probability of the user based on the service list and the logical change. In consideration of the path allocation state, it is possible to determine the optimum service list and the optimum logical path allocation state that maximize the network revenue and the network revenue.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a service list calculation device according to a first embodiment.
FIG. 2 is a block diagram illustrating a configuration example of a communication system.
FIG. 3 is a flowchart illustrating an operation of an optimum service list determination function unit.
FIG. 4 is a flowchart illustrating an operation of a target value maximizing service list selecting function unit.
FIG. 5 is a block diagram showing a modification of the service list calculation device.
FIG. 6 is a block diagram illustrating a configuration of a service list calculation device according to a second embodiment.
FIG. 7 is a schematic diagram showing one bottleneck model of a network.
FIG. 8 is a schematic diagram showing a network wide model.
FIG. 9 is a block diagram illustrating a configuration of a service list calculation device according to the related art.
[Explanation of symbols]
10. Optimal service list determination function section
11,12,13 Information acquisition unit
14 Selection probability calculation unit
15 Unit service network resource requirement calculation unit
16 Logical path target function value calculator
17 Optimal logical path selector
18 Target Value Maximization Service List Selection Department
21 History database
22 Calculation condition database
23 Demand forecasting function
24 Selection Probability Modeling Function
50 user terminals
100 network
101 Service list calculation device
102 Service list presentation system
103 User request reception system
104 QoS providing system

Claims (10)

When there are a plurality of service classes having mutually different communication service qualities, a plurality of service classes SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance, and a plurality of services are prepared. A service list selection device used to select a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s (i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition unit for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. Second information acquisition for acquiring a reference of a unit service network resource amount function r _j (SL) representing a network resource amount required per user of the service class j and a set of candidates that can be taken as a service list SL. Department and
A selection probability formula calculation unit that calculates the selection probability formula P ^s (j, SL) for the input service list SL;
A unit service network resource function calculating unit for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting bandwidth candidates and a finite number of routes that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and when each of the logical path candidates path ^{(i, e)} is selected, the expected revenue obtained by each logical path is targeted. A logical path target function value calculation unit that calculates a value Target (SL, path ^{(i, e)} );
For the input service list SL, 0-1 variables representing whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is calculated. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total revenue obtained from each logical path is defined as a target function value TargetValue, and the total revenue is formulated as a 0-1 integer programming problem with the maximization as an optimization condition. An optimal logical path selector for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. A process of calculating an optimum target function value and an optimum logical path by the optimum logical path selecting unit and updating the optimum target function value with the calculated value for all the service list SL candidates sequentially; A service list selecting apparatus, comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting unit for selecting an assignment state of a logical path. When there are a plurality of service classes having mutually different communication service qualities, a plurality of service classes SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance, and a plurality of services are prepared. A service list selection device used to select a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s ( i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability equation P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition unit for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. A reference of a unit service network resource amount function r _j (SL) representing a network resource amount required for one user of the service class j, a set of candidates that can be taken as a service list SL, and a per unit bandwidth of each link. A second information acquisition unit for acquiring cost information;
A selection probability formula calculation unit that calculates the selection probability formula P ^s (j, SL) for the input service list SL;
A unit service network resource function calculating unit for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting bandwidth candidates and a finite number of routes that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and using the cost information per unit band of each link, the cost for the set bandwidth in all the passing links is calculated as the expected expenditure, When each of the logical path candidates path ^{(i, e)} is selected, the expected profit obtained by subtracting the expected expenditure from the expected income obtained by each logical path is calculated as the target value Target (SL, path ^{(i, e)} ). Theory And pass target function value calculation unit,
For the input service list SL, 0-1 variables representing whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is calculated. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total return obtained from each logical path is defined as a target function value TargetValue, and the total return obtained from each logical path is formulated as a 0-1 integer programming problem with the optimization condition as an optimization condition, An optimal logical path selector for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. A process of calculating an optimum target function value and an optimum logical path by the optimum logical path selecting unit and updating the optimum target function value with the calculated value for all the service list SL candidates sequentially; A service list selecting apparatus, comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting unit for selecting an assignment state of a logical path. In the service list selecting device according to claim 1 or 2,
A history database that stores a history of past call occurrences and call durations and a history of service selection for the pair (i, e) of the network edge nodes;
A demand prediction unit that calculates the predicted call occurrence rate λs ^{(i, e)} and the predicted call termination rate ^{μs (i, e)} based on the history information read from the history database;
A user selection modeling unit that predicts a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL based on the history information read from the history database. Service list selecting device. When there are a plurality of service classes having mutually different communication service qualities, a plurality of service classes SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance, and a plurality of services are prepared. A service list selection method used to select a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s (i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition procedure for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. Second information acquisition for acquiring a reference of a unit service network resource amount function r _j (SL) representing a network resource amount required per user of the service class j and a set of candidates that can be taken as a service list SL. Instructions and
A selection probability equation calculation procedure for calculating the selection probability equation P ^s (j, SL) for the input service list SL;
A unit service network resource function calculation procedure for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting band candidates and a finite number of paths that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and when each of the logical path candidates path ^{(i, e)} is selected, the expected revenue obtained by each logical path is targeted. A logical path target function value calculation procedure calculated as a value Target (SL, path ^{(i, e)} );
For the input service list SL, 0-1 variables indicating whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is prepared. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total revenue obtained from each logical path is set as a target function value TargetValue, and is formulated as a 0-1 integer programming problem in which the maximization is an optimization condition. An optimal logical path selection procedure for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. Calculating the optimum target function value and the optimum logical path in the optimum logical path selection procedure and updating the optimum target function value with the calculated values for all the service list SL candidates sequentially; A service list selecting method, comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path. When there are a plurality of service classes having mutually different communication service qualities, a plurality of service classes SL representing a plurality of service classes that can be provided and a price assigned to each service class are prepared in advance, and a plurality of services are prepared. A service list selection method used to select a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s (i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition procedure for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. A reference of a unit service network resource amount function r _j (SL) representing a network resource amount required for one user of the service class j, a set of candidates that can be taken as a service list SL, and a per unit bandwidth of each link. A second information acquisition procedure for acquiring cost information;
A selection probability equation calculation procedure for calculating the selection probability equation P ^s (j, SL) for the input service list SL;
A unit service network resource function calculation procedure for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting bandwidth candidates and a finite number of routes that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and using the cost information per unit band of each link, the cost for the set bandwidth in all the passing links is calculated as the expected expenditure, When each of the logical path candidates path ^{(i, e)} is selected, the expected profit obtained by subtracting the expected expenditure from the expected income obtained by each logical path is calculated as the target value Target (SL, path ^{(i, e)} ). Theory And pass target function value calculation procedure,
For the input service list SL, 0-1 variables representing whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is calculated. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total return obtained from each logical path is defined as a target function value TargetValue, and the total return obtained from each logical path is formulated as a 0-1 integer programming problem with the optimization condition as an optimization condition, An optimal logical path selection procedure for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. A process of calculating an optimal target function value and an optimal logical path by the optimal logical path selection procedure and updating the optimal target function value with the calculated values for all service list SL candidates sequentially; A service list selecting method, comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path. In the service list selecting method according to claim 4 or 5,
The predicted call occurrence rate λ ^s based on history information read from a history database previously holding a history of past call occurrences and call durations for the pair of network edge nodes (i, e) and a history of service selection. ^{(I, e)} and a demand prediction procedure for calculating a predicted call termination rate μ ^{s (i, e)} ;
A user selection modeling procedure for predicting a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL based on the history information read from the history database. A service list selecting method. When there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing the plurality of service classes that can be provided and the price assigned to each service class are prepared in advance, and a plurality of service lists SL are prepared. In a computer-executable program for selecting a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s (i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition procedure for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. Second information acquisition for acquiring a reference of a unit service network resource amount function r _j (SL) representing a network resource amount required per user of the service class j and a set of candidates that can be taken as a service list SL. Instructions and
A selection probability equation calculation procedure for calculating the selection probability equation P ^s (j, SL) for the input service list SL;
A unit service network resource function calculation procedure for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting band candidates and a finite number of paths that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and when each of the logical path candidates path ^{(i, e)} is selected, the expected revenue obtained by each logical path is targeted. A logical path target function value calculation procedure calculated as a value Target (SL, path ^{(i, e)} );
For the input service list SL, 0-1 variables indicating whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is prepared. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total revenue obtained from each logical path is set as a target function value TargetValue, and is formulated as a 0-1 integer programming problem in which the maximization is an optimization condition. An optimal logical path selection procedure for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. Calculating the optimum target function value and the optimum logical path in the optimum logical path selection procedure and updating the optimum target function value with the calculated values for all the service list SL candidates sequentially; A program comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path. When there are a plurality of service classes having different communication service qualities, a plurality of types of service lists SL representing the plurality of service classes that can be provided and the price assigned to each service class are prepared in advance, and a plurality of service lists SL are prepared. In a computer-executable program for selecting a desired service list SL from the list SL,
When at least one of a division relating to a user, a division relating to an application to be used, and a division relating to content is represented as a segment s, a predicted call relating to a pair (i, e) of one network edge node i and another network edge node e. An occurrence rate λ ^{s (i, e)} and a predicted call termination rate μ ^{s (i, e),} and a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL. A first information acquisition procedure for acquiring
A network topology representing a network configuration, bandwidth information representing an available bandwidth of each link constituting the network, constraints on a bandwidth candidate and a route candidate of each logical path on the network, and a certain service list SL. A reference of a unit service network resource amount function r _j (SL) representing a network resource amount required for one user of the service class j, a set of candidates that can be taken as a service list SL, and a per unit bandwidth of each link. A second information acquisition procedure for acquiring cost information;
A selection probability equation calculation procedure for calculating the selection probability equation P ^s (j, SL) for the input service list SL;
A unit service network resource function calculation procedure for calculating the unit service network resource function r _j (SL) for the input service list SL;
For the input service list SL, calculation results of the predicted call occurrence rate λ ^{s (i, e)} , the predicted call termination rate μ ^{s (i, e),} and the selection probability formula P ^s (j, SL) And a calculation result of the unit service network resource amount function r _j (SL), a finite number of discrete value setting bandwidth candidates and a finite number of routes that meet the constraints on the bandwidth candidates and the route candidates of each logical path. A finite number of logical path candidates path ^{(i, e)} composed of combinations of candidates are determined, and using the cost information per unit band of each link, the cost for the set bandwidth in all the passing links is calculated as the expected expenditure, When each of the logical path candidates path ^{(i, e)} is selected, the expected profit obtained by subtracting the expected expenditure from the expected income obtained by each logical path is calculated as the target value Target (SL, path ^{(i, e)} ). Theory And pass target function value calculation procedure,
For the input service list SL, 0-1 variables representing whether or not each route is selected are prepared for all logical path candidates between all network end nodes, and the calculated target of each logical path candidate is calculated. Using the value Target (SL, path ^{(i, e)} ), the constraint condition that the total of the selected logical path setting bands passing through each link does not exceed the usable link band, and the selection from the logical path candidates Under the constraint that only one is performed, the total return obtained from each logical path is defined as a target function value TargetValue, and the total return obtained from each logical path is formulated as a 0-1 integer programming problem with the optimization condition as an optimization condition, An optimal logical path selection procedure for calculating an optimal target function value and an optimal logical path assignment when the service list SL is fixed by solving this problem;
For the first service list SL, the optimal target function value is initialized, and for each of the other service list SL candidates, the logical path candidate path ^{(i, e)} when the service list SL is fixed is used. Calculating the optimum target function value and the optimum logical path by the optimum logical path selection procedure and updating the optimum target function value with the calculated value for all the service list SL candidates sequentially; A program comprising: a service list SL in which a target function value TargetValue is finally maximized; and a target value maximizing service list selecting procedure for selecting an assignment state of a logical path. In the program according to claim 7 or 8,
The predicted call occurrence rate λ ^s based on history information read from a history database previously holding a history of past call occurrences and call durations for the pair of network edge nodes (i, e) and a history of service selection. ^{(I, e)} and a demand prediction procedure for calculating a predicted call termination rate μ ^{s (i, e)} ;
A user selection modeling procedure for predicting a selection probability formula P ^s (j, SL) representing a probability that the segment s selects the service class j from the service list SL based on the history information read from the history database. A program characterized by the following. 10. A computer-readable recording medium on which the program according to claim 7, 8, or 9 is recorded.

Images