JP3834258B2 - Quality state monitoring method and apparatus in packet network - Google Patents

Description

といった関数との合成関数である新たな系列ｈ_s ^c _n , ｈ^cp _n を得る。実際には、関数ｇを周波数ｆでサンプリングした系列ｇ_n を作成し、ｘ_s ^c _n , ｘ^cp _n とｇ_n の合成関数であるｈ_s ^c _n , ｈ^cp _n を、高速フーリェ変換（ＦＦＴ）を用いて算出する。
【００２０】
「相互相関演算機能」
本装置では、合成関数作成機能２１により合成された系列ｈ_s ^c _n , ｈ^cp _n （０≦ｃ≦Ｃ，１≦ｐ≦Ｎ）を受け取り、送信側データ系列ｈ_s ^c _n と一つの受信側データ系列ｈ^cp _n を対にして相互相関をＦＦＴによって求め、新たな系列ｒ^cp _n を得る。ここでは、Ｃ×Ｎ回の計算が行われる。なお、品質低下判定機能２３で必要となる場合には、ｈ_s ^c _n の自己相関をＦＦＴにより求め、新たな系列ｒ_s ^c _n を求める。なお、トラフィックを受動的にプローブしている際に、△Ｔが求められていない場合には、本機能においてｈ_s ^c _n を算出して△Ｔを求める。ここで、ｒ_s ^c _n , ｒ^cp _n に対して
［外２］

といった指数的なウェイトをかけた結果を新たなｒ_s ^c _n , ｒ^cp _n とする方法も考えられる。
【００２１】
「品質低下判定機能」
本装置では、相互相関演算機能２２より全てのデータ列ｒ_s ^c _n , ｒ^cp _n （０≦ｃ≦Ｃ，１≦ｐ≦Ｎ）を受け取り、０から△Ｔの範囲、もしくは、最初のピークが現れるラグをｔ^cpとし、ｔ^cpからｔ^cp＋△Ｔの範囲となるデータを抜き出し、これを新たなｒ_s ^c _n , ｒ^cp _n とする。データ点数は、ｍ＝△Ｔ／ｆとなる。
また、これらを用いて、以下のいずれかの方法により品質評価を行う。
・ｒ^cp _n の最大値と最小値を求め、その差の絶対値をＥ^cpとする。
・ｒ^cp _n の最大値を求め、その値をＥ^cpとする。
・
［外３］

となるＥを求め、
［外４］

となるＦ^cpを求める。次に、Ｅ^cpをＦ^cpで割ることによりＥ^cpを正規化する。
・０から△Ｔの範囲において、バスタブカーブとなる関数を周波数ｆでサンプリングした系列ｗ_m を用意し、
［外５］

となるＥを求め、
［外６］

となるＦ^cpを求める。次に、Ｅ^cpをＦ^cpで割ることによりＥ^cpを正規化する。
【００２２】
または、データ系列の範囲を抜き出さずに、以下のようにしてＥを求める。
ｒ^cp _n のパワースペクトルを求め、△Ｔの周期となる周波数成分のスペクトル密度をＥ^cpとする。なおｒ_s ^c _n のパワースペクトルを求め、△Ｔの周期となる周波数成分のスペクトル密度でＥ^cpを正規化する。
次に、Ｅ^cpが１に近いスレショルド値を超えない場合には、１^cp＝０とし、スレショルドを超えた場合には１^cp＝１とする。優先度誤りを監視する場合には、受信側パケットプローブ番号において、Ｅ^0P＜Ｅ^1P＜Ｅ^2P＜・・・＜Ｅ^NPであることを確認する。この関係が保たれている場合には、品質低下判定結果１ついて、１^P ＝１とする。これはクラス間での品質低下が発生していないことを示す。上記関係が保たれていない場合には、１^P ＝０とする。これは、クラス間での優先度誤りが発生していることを示す。なお、以降では、ある特定のクラスにおける１^cpについても１^P と表記する。
【００２３】
次にこの操作を全てのＥに対して行い、１^P と、Ｅ^cpもしくはｈ^cp _n もしくはｒ^cp _n を出力値とする。
以上の操作を連続して行うことにより常時監視が実現される。
【００２４】
（トポロジー把握装置）
トポロジー把握装置３０は、経路情報推定機能３３、経路情報収集機能３４、分岐合流短絡除去機能３１、及び論理トポロジー作成機能３２により実現される。経路情報推定機能３３と経路情報収集機能３４は、従来技術により実現可能である。例えば、経路情報推定機能３３は、受信側パケットロープにおいてパケットロスを観測することにより行うことができ、または、パケット伝播経路を管理している場合にはその管理情報を用いて、または、ＳＮＭＰ（Simple Network Management Protocol）といったネットワーク管理プロトコルを利用することにより実現可能である。従って、パケットプローブ装置からパケットロス情報が経路情報推定機能３３に供給され、当該機能３３により推定された経路情報が論理トポロジー作成機能３２に供給される。ここでは、これらの技術によりパケット転送経路が得られたものとして説明を続ける。なお、パケット転送経路は、一台の送信機から受信機までの、ループのない経路となる。
【００２５】
経路情報収集機能３４は、パケット伝搬経路を管理している場合の管理情報又はネットワーク管理プロトコルを利用することにより経路情報を収集し、収集された経路情報を分岐合流短絡除去機能３１及び論理トポロジー作成機能３２に供給する。
【００２６】
分岐合流短絡除去機能３１、論理トポロジー作成機能３２では、経路情報推定機能３３もしくは経路情報収集機能３４により得られたパケット転送経路を、品質低下個所判定装置４０において利用可能な論理トポロジーに変換する機能を有する。論理トポロジーは木構造であり、ノードＶ、リンクＬ、ルートＶｒ、リーフＶｌにより構成される。リーフＶｌは回線とルータやスイッチの集合を示し、通常ルータやスイッチからなる経路の分岐点を示す。リンクＬは回線とルータにより構成される分岐のないパケット転送経路を示す。ルートＶｒは送信側パケットプローブを示し、リーフＶｌは受信側パケットプローブを示す。これらは以下の規則に従って接続されている。
・Ｖｒは、１個以上のノードＶが、もしくは、一個以上のリーフＶ１が、もしくは、一個以上のノードＶと一個以上のリーフＶ１が、リンクＬにより接続されている。
・リーフＶ１は、２個以上のノードＶが、もしくは、2 個以上のリーフＶ１が、もしくは、一個以上のノードＶと１個以上のリーフＶ１が、リンクＬにより接続されている。
・リーフＶ１は、１個のノードＶもしくは１個のルートＶｒにリンクＬにより接続されている。
【００２７】
実際のパケット転送経路は、一台のパケット送信機から見た場合に、その経路を有向グラフとすると、全てのルータやスイッチでは単方向連結するように経路設定されており．強連結成分は存在しない。しかし、分岐した後に合流する経路を持つ構造となる可能性がある。このため、本発明では、上記のルールに従った木構造となる論理トポロジーを得るために、第一段階として分岐合流短絡除去機能により、分岐して合流する経路を発見して、分岐と合流で閉じた領域にあるリンクとノードを全て短絡除去し、木構造となるトポロジーを得る作業を行う。なお、マルチキャストを用いて測定を行う場合には、分岐合流短絡除去機能を利用せず、経路情報推定機能３３もしくは経路情報収集機能３４から、直接論理トポロジー作成機能にマルチキャスト木を出力することも可能である。強連結成分（ループ）の存在は経路制御の問題であり、本発明では経路制御が正常に行われて強連結成分がないことを前提としている。
【００２８】
「分岐合流短絡除去機能」
本装置では、分岐と合流で閉じた領域にあるリンクを全て短絡除去する。短絡除去を行う一つの方法として、以下ではリンクを２色に色分けする方法による分岐合流短絡除去機能の一実施例を示す。
伝播トークンは、１個以上のデータトークンを持つ。データトクンは、分岐数、分岐ノード番号、色、及び始発リーフ番号の組からなるデータを持つ。またデータトークンは、分岐数、分岐ノード番号、色、及び始発リーフ番号の組からなるデータのスタック領域を持つ。また、パケット転送経路のノードは重複しないノード番号を持つ。この際に、トークンを以下のルールに従って流す。
・一つのリンク上では、一つの伝播トークンのみが有向木（経路テーブル）の方向と逆向きに通過できる。
・伝播トークンがノードに到着した際には、伝播トークンから全てのデータトークンが取り出される。
・ノードから出力される場合には、全てのデータトークンが伝播トークンに包まれる。
・データトークンの初期の属性は、分岐数＝なし、分岐ノード番号＝なし、色＝白色、スタックは空であり、トークンが生成されたリーフ番号が始発リーフ番号として与えられている。また、リンクの初期属性は白色である。
・伝播トークンが分岐、すなわち、２本以上の入力を持つノードにたどりついた際には、伝播トークンからデータトークンを取り出し、全てのデータトクンについて、現在の属性をスタックに格納し、分岐数＝分岐リンク数、色＝赤色、分岐ノード番号＝ノード番号を新たな属性として持たせる。始発リーフ番号は変更しない。次に、全てのデータトークンを一つの伝播トークンで包み、伝播トークンのコピーを作成し、同一伝播トークンを全てのリンク上に流す。
・リーフから複数に分岐している場合には、上記と同様の属性を持つデータトクンを生成して、そのデータトクンを一つの伝播トークンで包み、伝播トークンのコピーを作成し、同一伝播トークンを全てのリンク上に流す。
・赤色属性を持つデータトークンを含む伝播トークンが流れたリンクでは、リンクの属性が白色の場合には赤色と変更する。リンクの属性は赤色から白色となることはない。
・トークンが合流、すなわち、２本以上の出力を持つノードにたどり着いた場合には、全てのリンクから伝播トクーンが到着するまで待機して、以下の動作を行う。
同一の分岐ノード番号を持ち、かつ、同一の始発リーフ番号と分岐ノード番号を持つデータトークンについて、データトークンの個数が，データトークンの属性である分岐数と等しい場合には、データトークンの属性をスタックから取り出した属性と入れ替えて一つのトークンにまとめる。この作業を可能な限り繰り返し行う。
まとめるデータトークンがない場合には、全てのデータトークンを一つの伝播トークンで包んで出力する。
【００２９】
全てのリーフから上記のルールに従ってトークンを渡すことによって、リンクを色分けすることができる。このようにして色分けされたリンクの中で、赤色部分について短絡除去を行い、一つのノードとすることにより、木構造となる論理トポロジーを得ることが出来る。
【００３０】
図３に動作例を示す。図３において、太い実線は赤色属性を持つリンクを示し、細い実線は白色属性を持つリンクを示し、矢印の向きは経路テーブルが指す方向を示す。丸はノードであり、ノード番号が丸の中に示されている。四角はデータトークンであり，（ａ，ｂ，ｃ，ｄ）は属性を示し，［（ａ，ｂ，ｃ，ｄ）］はスタックに格納された属性を示す。ここで、ａは分岐数を示し、ｂは分岐ノード番号を示し、ｃは色属性を示し、ｄは始発リーフ番号を示す。スタックでは、属性は上に詰まれるように格納され上から順に取り出される。吹き出しは伝播トークンを示す。
【００３１】
図３では、トークンが始点リーフ番号１から出されたとして例示している。初期時は、全てのリンクの属性は白色であり、伝播トークンの中には一つのデータトクンがあり、（−，−，白，１）の属性を持つ。この伝播トークンが有向グラフ上を逆向きに流れている。ノード６において２本に分岐されるため、データトクンは、（２，６，赤，１）の属性となり、（−，−，白，１）の属性はスタックに格納される。赤色属性である、データトークンを持つ伝播トークンが通過したため、リンクの属性が赤色となる。次に、ノード５において、再度分岐が行われるため、データトークンは、（２，５，赤，１）の属性となり、（２，６，赤，１）の属性はスタックに格納される。ノード２において合流が行われるが、まとめることが可能なデータトークンが存在しないため、全てのデータトークンを一つの伝播トークンにまとめて上流に渡す。ノード１において合流が行われる。まず、（２，５，赤，１）の属性を持つデータトークンが２個現れるため、一つのデータトークンとしてまとめ、スタックから取り出した（２，６，赤）となる属性を与える。さらに、ノード１では、（２，６，赤，１）の属性を持つデータトークンが２個となるため、これらもまとめて、一つのデータトークンとする。この結果、図の最後のように、合流、分岐箇所が赤色リンクとして表現され、このリンクを短絡除去することにより、２本のリンクと一個のノードに変更することが出来る。
【００３２】
「論理トポロジー作成機能」
本装置では分岐合流短絡除去機能により得られた木構造をもとにして、前述した規則に従った構造となるように変更を行う。本装置では、一本のリンク入力と持ち、一本のリンクを出力と持つノードとリンクからなる部分を一本のリンクに変更する。図４に実際の変更を例示する。図４において、丸はノードを示し、矢印はリンクを示し、四角はリーフを示している。さらに、短絡するノードについて斜線で示してある。以上の▲１▼▲２▼に示した装置により論理トポロジーが作成され、トポロジー把握装置の出力となる。
【００３３】
（品質低下個所判定装置）
品質低下個所判定装置４０は、品質監視装置２０からＩ^P を受け取り、低下箇所判定機能４１と低下箇所推定機能４２によって品質低下個所を推定する機能である。以下ではこれらの装置の処理順序に従って説明を行う。
「低下箇所判定機能」
品質監視装置２０が品質低下の発生を判断した際に、低下箇所判定機能４１は品質監視装置２０から品質監視結果を受け取り、トポロジー把握装置３０から論理トポロジーを受け取る。低下箇所判定機能はこれらの結果を用いて処理を行う。
【００３４】
図５を用いて、基本的な判定方法を説明する。低下箇所判定機能は、品質監視装置からＩ^P を受け取り、低下箇所判定を行う。図５の左側に示すトポロジー情報と品質監視結果を受け取った場合に、ルートＶｒからリーフＶｌ２、Ｖｌ３への経路では、送信、受信パターンに違いが少なく、すなわち、パケットロスがなく、遅延変動が少ない品質であるといえる。しかし、リーフＶｌ１への経路上において、品質低下が発生している。ここでは、リーフＶｌ１への経路において、リーフ２、３への経路以外の箇所において、もしくは、リーフ２、３への経路とリーフ１への経路の分岐点での出力インターフェース部分において品質低下が発生していると考えられる。そこで、リーフ１への経路の中で、リーフ２、３への経路に含まれないリンクにマークをつけ、さらに、リーフ２、３への経路からリーフ１への経路の分岐点となるノードにマークを付ける。このマークの付いた箇所を品質低下が発生した可能性が高い箇所として出力する。図５において、品質低下が発生したリンクにマークを付して明瞭にする。このように、低下箇所判定機能４１は、品質監視装置から供給された品質低下判定結果とトポロジー把握装置３０から供給された論理トポロジーとに基づいて品質低下が生じたリンク部分を特定する。
【００３５】
以降では、論理トポロジーの構造に従って詳細な説明を行う。トポロジー把握装置から受け取ったトポロジー情報と、品質監視装置から受け取った一つのクラスに関する情報、もしくは、優先度誤りの判定結果を図６に例示する。図６において、リーフ内部の数字はリーフ番号を、括弧でくくられた内容はリーフの属性を示す。（○）は品質低下が発生していないこと、すなわち、Ｉ^P ＝１であり、（×）は品質低下が発生したこと、すなわち、Ｉ^P ＝０であることを示す。丸はノードを示し、ノード内部の数字はノード番号を示す。
【００３６】
最初に、リーフのみが接続しているノードにおいて、以下のルールに従って属性を付与する。図６では、ノード７，９，１０，１１，１３，１４，１５，１６，１７が対象となる。
・リーフの属性が全て（○）の場合には。ノードに（○，○）の属性を与える。
・リーフの属性が全て（×）の場合には。ノードに（×，×）の属性を与える。
・リーフの属性が（○）と（×）の場合には。ノードに（×，○）の属性を与える。
【００３７】
この結果を、図７に例示する。なお、（Ａ、Ｂ）の表記については、Ａはそのノードにおける品質状態を示し、これを自ノード状態と呼ぶ。Ｂはそのノードから上位部分において品質低下が発生した可能性を示し、これを伝播状態と呼ぶ。
【００３８】
次に、属性を持つノードとリーフのみを子としてもつノードについて、以下に従って属性を付与する。図では６、８、12が対象となる。
・全ての子ノードの伝播状態が○であり、子にリーフが存在する場合には全ての子リーフの属性が○の場合に、（〇，〇）とする。
・全ての子ノードの伝播状態が×であり、子にリーフが存在する場合には全ての子リーフの属性が×の場合に。（×，×）とする。
・子にリーフが存在する場合にはリーフの属性も含めて、子ノードの伝播状態に○と×が混在する場合に、（×，○）とする。
【００３９】
なお、ルートＶｒの属性も上記と同様とする。上記操作を繰り返し行った結果を、図８に例示する。
【００４０】
自ノード状態が×の場合には、下位ノード、もしくは、リーフにおいて品質低下が観測できることを示す。これは、上位ノードから既に品質低下が発生している可能性と、自ノードを含め自ノードから下位ノードの間で発生している可能性の二つが考えられる。しかし、自ノードの伝播状態が○である場合には、下位ノードの一部で品質低下が発生していない箇所があることを示すため、上位ノードから自ノードまでの区間において品質低下が発生している可能性が低いと考えられる。そこで、自ノードを含め自ノードから伝播状態が×となる子ノードヘのリンク、もしくは、（×）の属性を持つリーフヘのリンクで品質低下が発生した可能性が高いと考える。
【００４１】
前段落の仮定をもとにして、以下のルールに従ってマークを付与し、このマークの付いたノードとリンクが示すスイッチやルータの集合が、品質低下を起こした可能性のある箇所として低下箇所判定機能の出力とする。
・ルートも含めて自ノード状態が×となるノードにマークを付与する。
・マークが付与された個々のノードから伝播状態が×となる子ノードヘのリンク、もしくは、（×）の属性を持つリーフヘのリンクにマークを付ける。
【００４２】
この結果を図９に示す。なお、ルートの属性が（×，×）の場合には、ルートから見た子ノードに、もしくは、リーフにマークを付与し、かつ、そのノードもしくはリーフヘのリンクにマークを付ける。なお、図９において、ノード番号１，３，６，７，１５，１４の部分木内では、どの区間において品質低下が発生しているか判定できない。そこで、このような部分木について解析の必要があれば、品質低下が発生した可能性がある箇所を低下箇所推定機能４１により推定する。
【００４３】
「低下箇所推定機能」
（×，×）属性を持つノードやリーフが部分木を構成している場合には、その部分木を全て取り出す。以降では、取り出された部分木の一つについて推定方法を示すが、低下箇所推定機能４２は取り出された全ての部分木について以下の処理を行う。品質低下箇所を推定するための方法として、部分木内部の全てのリーフに属するデータ系列Ｅ^CPの履歴データ（過去24時間164 時間といった時系列データ）や、ｈ^CP _n もしくはｒ^CP _n を用いて相関を求めて低下個所を推定する方法が考えられる。簡単な説明を図１０を用いて行う。
【００４４】
図１０において、上下とも相関関係からリーフＶｌ２とＶｌ３には、同一のパケット時刻列が到着していることがわかる。しかし、リーフＶｌ１にはリーフＶｌ２及び３とは異なるパターンでパケットが到着している。品質低下個所を通過した後の経路において、品質低下個所がない場合には、Ｅ^CP、ｈ^CP _n 、ｒ^CP _n といったデータ系列間の相関関係が高くなると考えられる。このため、図の上については、ルートＶｒからリーフ１へ至る経路上とルートＶｒからノード２に至る経路上において、品質低下が発生している可能性があると考えられる。図の下については、ルートＶｒからリーフＶｌ１へ至る経路上に品質低下が発生している可能性があると考えられる。図１０では、このようにして求めた推定箇所にマークを付与している。低下箇所推定機能は、マークが付与された箇所を推定箇所として出力する。以降では、新たに図１１で示す木構造を用いて詳細に説明を行う。
【００４５】
木の中で、リーフと、リーフのみを子として持つノードとで構成される部分木を一つのグループとする。図１２にグループ化の結果を示す。図１２において、点線で囲まれた部分木がグループとなる。個々のグループにおいて、リーフに対応するデータ系列Ｅ^CPの履歴データ（過去２４時間、１６４時間といった時系列データ）、もしくは、ｈ^CP _n 、もしくは、ｒ^CP _n を用いて、グループ内のリーフの全組み合わせで相関の強弱を求める。計算の結果。相関が高い組み合わせは「〇」とし、低い組み合わせは「×」とする。この結果を利用して、グループ内のノードとリンクに対して以下のルールに従って属性を付与する。
・グループ内の全組み合わせが「〇」の場合には、グループ内の全てのノードとリンクに（○）の属性を付与する。
・グループ内の全組み合わせが「×」の場合には、グループ内の全てのノードとリンクに（×）の属性を付与す
・組み合わせの属性に「〇」と「×」が混在する場合には、ノードに（×）の属性を付与し、さらに、「〇」の組み合わせに対応する２つのリーフに至る２本のリンクに（○）の属性を付与し、それ以外のリンクに（×）の属性を付与する。
【００４６】
この結果を図１３に示す。なお、相関の計算数を減らすための一方法として、計算対象となるリーフを頂点に持つ多角形を作成し、フルメッシュ状を引き、三角形に注目して相関を求める方法が考えられる。例えば、ａ，ｂ，ｃ，ｄ，ｅからなる５つのリーフがある場合に、五角形の頂点をそれぞれａ，ｂ，ｃ，ｄ，ｅとし、ａ，ｂ，ｃの三角形に注目した祭に、ａ−ｂは〇と計算され、ｂ−ｃは×と計算された場合には、ａ−ｃが×となる。ｃ−ｄが×と計算された場合には、ａ−ｃは計算する必要があり、ａ−ｅが○と計算された場合には、b −e は○となる。これらは以下のようにまとめられる。
【００４７】
・２辺が○の場合には、３辺目は○とする。
・１辺が○であり、他の１辺が×の場合には。３辺目は×とする。
・2 辺が×の場合には、3 辺目は計算する。
【００４８】
次に、以下のように、代表データ系列群を選択し、ノードに与える。
・グループ内の全組み合わせが「〇」の場合には、任意に一つのリーフに対応するデータを代表データ系列群とする。
・グループ内の全組み合わせが「×」の場合には、全てのリーフに対応するデータを代表データ系列群とする。
・組み合わせに「〇」と「×」が混在する場合には「〇」となる組み合わせに注目し、関連のある（閉じた）組み合わせから相関の最も高い組み合わせを選択してデータを任意に一つ選び出し、他は対象外とする。例えば、図１３においては、ノード８の属するグループにおいて閉じた組み合わせはリーフＶｌ５，６，７であり、例えば、リーフＶｌ５のデータを代表データ系列群へ含む。また、ノード１０においては、リーフＶｌ１０と１１，１２と２２の二つがそれぞれ独立に閉じた組み合わせとなるため、例えば、リーフ１０と１２のデータを代表データ系列群へ含む。残りの「×」となる組み合わせの全リーフに対応するデータも代表データ系列群へ含む。この結果を図１４に示す。
【００４９】
次に、子に属性が付与されたノードと前段で演算対象とされていないリーフを持ち、自らは属性を付与されていないノードを処理対象ノードとする。これらのノードとリーフにより構成される部分木を新たなグループとする。図１４を例とすると、ノード４，１１が処理対象ノードとなり、グループは点線で囲まれている。
【００５０】
個々のグループにおいて、処理対象ノードの子ノードが持つ代表データ系列群について相関を求める。この際に、代表データ系列群は、処理対象ノードから見た子ノードに属する単位でまとめられ、これを部分代表データ系列群と呼ぶ。子にり−フがある場合には、リーフに対応するデータ系列を一つの部分代表データ系列群とする。
【００５１】
処理対象ノードでは、部分代表データ系列群間で相互相関を計算する。部分代表データ系列群内のデータ系列間では組み合わせない。例えば、処理対象ノードに子ノードＡ，Ｂが存在し、ノードＡの代表データ系列群が1 ，2 であり、ノードＢの代表データ系列群が３，４である場合には、１，２と３，４という二組の部分代表データ系列群にまとめられ。組み合わせは１−３、１−４、２−３、２−４となる。
【００５２】
部分代表データ系列群間での相関の計算の結果、一つでも「○」となる組み合わせがある場合には、部分代表データ系列群間の相関は（○）とする。全ての組み合わせが「×」の場合には、部分代表データ系列群間の相関は（×）とする。
求めた相関の結果を用いて以下のようにノードとリンクに属性を与える。
・部分代表データ系列群間の相関が全て（○）の場合には、グループ内の処理対象ノードとリンクに対して「○」の属性を与える。
・部分代表データ系列群間の相関が全て（×）の場合には、グループ内のノードとリンクに対して「×」の属性を与える。
・部分代表データ系列群間の相関に、（○）と（×）が混在する場合には、（○）となる部分代表データ系列に対応する子ノードと子ノードへ至るリンクに「○」の属性を付与し、残りの子ノードとリンクに「×」の属性を与える。また、処理対象ノードに「×」の属性を与える。
【００５３】
次に、代表データ系列群を選定する。この結果を図１５に示す。
・グループ内の組み合わせが「〇」の場合には、相関の最も高い組み合わせの中から任意に一つのデータを代表データ系列群とする。
・グループ内の全組み合わせが「×」の場合には、全てのデータを代表データ系列群とする。
・組み合わせに「〇」と「×」が混在する場合には、「〇」となる組み合わせの中に注目し関連のある（閉じた）組み合わせから相関の最も高い組み合わせを選択してデータを任意に一つ選び出し、他は対象外とする。例えば、図１５においては、部分代表データ系列７と８の間において、２−５と４−９という２種類の閉じだ組み合わせがある。この場合は例えば、２，４を代表データとし、５，９を対象外とする。残りの「×」となる組み合わせの中で対象外とならなかったデータも代表データ系列群とする。
【００５４】
この結果を図１６に例示する。この後、新たな処理対象ノードを決定して同様の計算を行う。以上の操作を処理対象ノードがなくなるまで行う。最後に（Ｘ）の属性を持つノードとリンクにマークを付ける。実際の表示例を図１７に示す。
【００５５】
【発明の効果】
本発明により、優先度制御を行ったネットワークの品質制御の状態を監視し、優先度誤りといった制御異常状態が把握可能になる。また、品質異常が発生した際に、異常箇所を把握することができる。品質測定時に片道遅延時間といった時刻に関する品質測定を不要とすることができる。これにより、複数箇所に分散した測定装置が周波数同期のみを実現することにより、品質が可能になる。
【図面の簡単な説明】
【図１】 パケット網における品質状態監視装置の構成図である。
【図２】 パケット網における品質状態監視装置の実施例である。
【図３】 パケット経路情報から分岐、合流の構造を短終除去する一実現方法例である。
【図４】 短縮除去したトポロジーより論理トポロジーを作成するための変更例である。
【図５】 品質低下箇所把握の基本的な把握方法の例示である。
【図６】 品質監視装置とトポロジー把握装置の出力結果例である。
【図７】 リーフの属性よりノードの属性を作成した結果の例示である。
【図８】 子ノードの属性から親ノードの属性を作成した結果の例示である。
【図９】 品質低下が発生していると判定した箇所の例示である。
【図１０】 相関関係を基にした簡単な推定方法の例示である。
【図１１】 低下箇所推定機能が処理を行う有向木例である。
【図１２】 ノードとリーフのグループ化の例である。
【図１３】 グループ化された内部の処理結果の例示である。
【図１４】 代表データ系列群の選定結果の例示である。
【図１５】 低下箇所推定機能の処理結果の例示である。
【図１６】 代表データ系列群の選定結果の例示である。
【図１７】 最終的な品質低下箇所推定結果の例示である。
【符号の説明】
１ パケット網
１１，１２ パケットプローブ装置
２０ 品質監視装置
２１ 合成関数作成機能
２２ 相互関数演算機能
２３ 品質低下判定機能
３０ トポロジー把握機能
３１ 分流合流短絡除去機能
３２ 論理トポロジー作成機能
３３ 経路情報推定機能
３４ 経路情報収集機能
４０ 品質低下箇所判定装置
４１ 低下箇所判定機能
５０ 表示装置[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a packet network quality state monitoring method for monitoring a quality state of a packet network and determining a part of the quality information when the quality information is found to be deteriorated. In particular, the present invention relates to a quality state monitoring method suitable for monitoring priority transfer control in a network in which QoS control is performed in which packet transfer priority is classified as a transfer class and priority transfer control is performed for each class. It is about.
Furthermore, the present invention relates to a quality state monitoring method suitable for monitoring packet transfer quality of real-time applications that are easily affected by delay fluctuations and band fluctuations such as video and audio transfer.
[0002]
[Prior art]
In order to realize quality control monitoring by priority control by class, it can be realized by comparing and monitoring the state of delay variation between classes. Further, in order to transfer a real-time application with high quality, it is required that the delay variation is small. For this reason, in order to realize quality monitoring, a method of calculating the one-way delay time as follows can be considered.
-Perform frequency synchronization and time synchronization in the packet probe device.
• Probe packets between packet probe devices.
Receive a probe data sequence with a probe time from the packet probe devices on the transmission side and the reception side.
Calculate the one-way delay time from the probe time of the detected packet to obtain the distribution.
[0003]
Next, as another method for quality monitoring, a method for calculating jitter indicating delay variation is conceivable. In order to obtain jitter, the following processing must be performed.
• Perform frequency synchronization in the packet probe device.
• Probe packets between packet probe devices.
Receives probe data strings with probe times from the packet probe devices on the transmission side and the reception side.
Search for the same packet from these data strings.
・ Calculate the one-way delay time from the probe time between transmission and reception.
・ Jitter is calculated from the one-way delay time to obtain the distribution.
[0004]
As described above, in order to measure the one-way delay time, it is necessary to calculate the one-way delay time by obtaining the difference between the packet transmission time and the reception time. For this reason, in the measurement of the one-way delay time, it is required that the time of the clock is advanced (frequency synchronization) after all the times of the packet probe devices are matched (time synchronization).
[0005]
The method of measuring quality by measuring jitter does not require accurate time synchronization because jitter is determined as a one-way delay time difference. Therefore, in order to obtain the jitter distribution and perform quality control monitoring, it is only necessary to realize frequency synchronization of the packet probe apparatus. In order to obtain the one-way delay time and jitter, it is necessary to perform the same packet search the number of times corresponding to the number of probe packets from a set of data strings on the transmission side and the reception side. Further, in the method for obtaining the one-way delay and jitter distribution, packet loss resulting from quality fluctuation is not taken into consideration.
[0006]
Next, in order to perform quality control monitoring by the method of obtaining the one-way delay time and jitter, and to point out the location when the quality deteriorates, packet probe devices are exhaustively arranged inside the monitored network, and between each probe device The method of grasping the quality degradation section from the measurement result in can be considered. In this method, a measurement interval is defined between two pairs of probe devices, a one-way delay time and jitter are calculated within the measurement interval, and the presence or absence of quality degradation is determined from the result. If quality degradation has occurred, report the measurement interval as a quality degradation location. By performing this process in all the packet probe devices, it is possible to determine a location where quality has deteriorated.
[0007]
[Problems to be solved by the invention]
As described above, accurate time synchronization and frequency synchronization are required for measuring the one-way delay time. However, in recent years, when the bandwidth of a network has advanced and there is no queuing delay, a minimum delay time of several microseconds / km is provided. For this reason, time synchronization for obtaining the one-way delay time requires an accuracy of the order of microseconds or less. In order to achieve such accuracy, a method using GPS is conceivable, but there is a problem that the installation location is limited, and it is difficult to exhaustively distribute packet probes. Further, a clock synchronization method using ISDN has been proposed, but with this method, time cannot be synchronized with an accuracy of ± 10 μs or less.
[0008]
However, frequency synchronization can be realized if there is an accurate frequency source such as frequency synchronization using ISDN. Frequency synchronization can achieve an accuracy of a few microseconds or less. For example, since the time for which a 1500-byte packet passes on a 1 Gbit / s network is 12 μs, 10^--6 It can be measured with the accuracy of seconds. Therefore, a measurement method that can be realized only by frequency synchronization, that is, quality monitoring by obtaining jitter can be considered.
[0009]
Here, it is assumed that a failure occurs in a network in which priority control is performed as a quality control method, in which the priority of packets set to a high priority is equal to the priority of other traffic set to a low priority. In this situation, if there is a large amount of non-measurement packet traffic, there is a possibility that the test packet is always located at the end of the queue. In this case, the delay time becomes long, but the delay variation may be small. For this reason, in such a state, if a method using a jitter distribution is applied, no quality degradation can be found. In order to avoid such a problem, considering that the occurrence probability of packet loss is very high when a packet is always located behind queue, there is an element of packet loss in the quality monitoring method. It is necessary to select the method to be included.
[0010]
Next, the problem is described about the grasp method of the quality degradation location. It is necessary to arrange packet probe devices exhaustively in order to grasp the quality-decreasing location after the failure is discovered. At this time, it is necessary to receive probe data strings from all the packet probe devices and perform the same packet search from the data strings at both ends in the set monitoring section. When packet probe devices are arranged exhaustively, an increase in one packet probe device results in an increase in one or more monitoring intervals, which increases the number of tables for searching for the same packet required for jitter calculation. Is an issue. Furthermore, when the delay variation is large, it is necessary to enlarge the table for searching for the same packet, and the time required for searching for the same packet may become very long.
[0011]
OBJECT OF THE INVENTION
An object of the present invention is to use only an accurate frequency source without using high-precision time synchronization, include packet loss information in addition to delay variation information, and perform quality monitoring without performing the same packet search. It is to provide a quality monitoring method.
Furthermore, another object of the present invention is to provide a method for determining a quality degradation location where quality degradation has occurred using end-end measurement without exhaustively arranging packet probe devices in the network. That is.
[0012]
[Means for Solving the Problems]
FIG. 1 shows a basic configuration of a quality state monitoring system including a quality state monitoring device in a packet network according to the present invention. The quality time monitoring device according to the present invention includes a quality monitoring device 20 that monitors the quality state of the packet network 1, a topology grasping device 30, and a quality degradation point determination device 40. As shown in FIG. 1, a plurality of packet probe devices 11 to 13 driven by a high-precision frequency source and having a time string generation function are installed at the end of a network 1 to be measured and monitored. The packet probe device probes the packet and performs end-to-end measurement. At this time, the packet transmission time sequence is obtained in the transmission side packet probe device 11 by the time sequence generation function of each packet probe device, and the packet reception time sequence is obtained in each of the reception side packet probe devices 12 and 13.
[0013]
The transmission time sequence obtained by the transmission side packet probe device 11 and the packet reception time sequence obtained by the reception side packet probe devices 12 and 13 are supplied to the quality monitoring device 20. The quality monitoring device 20 obtains a cross-correlation function from the packet transmission time sequence and the packet reception time sequence obtained as a result of the end-to-end measurement using the synthesis function creation function 21 and the cross-correlation calculation function 22. Based on the obtained cross-correlation function, the quality degradation determination function 23 determines the presence or absence of quality degradation and outputs a quality degradation judgment result. Thereby, quality monitoring is realized without performing the same packet search as that of high-precision clock synchronization. In addition, when the difference in priority is observed as packet loss, the correlation becomes low. For example, it is possible to measure a deterioration in quality such that the packet is always located behind the queue.
[0014]
The topology grasping device 30 uses the branch / junction / short-circuit removal function 31 to remove the branch / junction of the route information by short-circuiting from the packet transfer route information collected by the network management information or the like. Next, the logical topology creation function 32 removes redundant structures and creates a logical topology having a tree structure. Here, the root of the tree is a transmission side packet probe device, and the leaf is all the reception side packet probe devices. The direction indicates the packet transfer direction. Using this logical topology, if quality degradation has occurred, the location where quality degradation has occurred is determined.
[0015]
In the quality degradation location determination device 40, the quality degradation determination result and data series are received from the quality monitoring device 20, the ethical topology is received from the topology understanding device 30, and the quality degradation has occurred by the degradation location determination function 41. To grasp. Next, the lowering portion estimation function 42 estimates the lowering portion more finely. And the output from the fall location determination function 41 and the fall location estimation function 42 can be visualized and displayed on the display device 50.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 2 shows a processing flow of the quality monitoring apparatus for a packet network according to the present invention as an embodiment of the present invention. In addition, the same code | symbol is attached | subjected and demonstrated to the component same as the component used in FIG. In FIG. 2, a portion indicated by a thick solid line corresponds to the present invention. In the following, these functions will be described in detail.
(Packet probe device)
In the packet probe device, a timepiece is operated using a high-accuracy frequency source, and the way of the timepiece is accurately synchronized. This apparatus has a time string generation function. This function is described below.
"Time sequence generation function"
In this function, the packet passage time is converted into a data sequence sampled at the sampling frequency f. At this time, the time when the packet was probed is set to 1, and the others are set to 0 to generate a data string. For example, when f = 10 Hz and a test packet is probed at 0.0 seconds and 0.3 seconds, 0.0 seconds = 1, 0.1 seconds = 0, 0.2 = 0, 0.3 A data series is generated as seconds = 1. This data string is divided by the time interval t and is continuously transmitted to the quality monitoring apparatus 20. Using the above example, when t = 1 second, 10 samples are transmitted to the quality monitoring apparatus 20 every second.
[0017]
The data series created from the sender time column is x_s ^c _n And the data series created from the receiving time sequence is x^cp _n Represent as Here, c indicates a quality control class to which the data string belongs, and is indicated as 1, 2, 3,..., C in order from 0, and for example, the priority increases from 1 to c. Numbers 1, 2, 3,..., N are assigned to the reception side packet probe devices in order from 1, and p indicates a reception side packet probe device number. In the case of a network that is not classified, C has only one fixed value.
[0018]
When the quality is dynamically evaluated using a test packet, a test packet is sent from one packet probe device to all other packet probe devices using multicast or a plurality of unicasts. The test packet is formed as a continuous constant pattern having periodic characteristics at a constant interval ΔT. The packet probe device pays attention to the test packet generated in this way and collects the transmission or reception time of the test packet. When passively evaluating the quality of traffic flowing in end-to-end, a packet probe device at an end point between transmission and reception probes only the traffic and collects transmission time and reception time. Note that the periodicity of the packet that has been transferred in advance is measured, and the value is taken as ΔT. For example, when packets are generated at regular intervals, the interval is set to ΔT.
[0019]
(Quality monitoring device)
  A quality monitoring device 20 that monitors the quality of the network is realized by a synthesis function creation function 21, a cross-correlation calculation function 22, and a quality degradation determination function 23. Hereinafter, the operation contents will be described according to the processing flow of these apparatuses.
"Synthetic function creation function"
  In this apparatus, the above-mentioned transmission side and reception side data series x_s ^c _n And x^cp _n For (0 ≦ c ≦ C, 1 ≦ p ≦ N), a function g prepared in advance, for example,
[Outside 1]

A new sequence h that is a composite function with a function such as_s ^c _n , h^cp _n Get. Actually, the sequence g obtained by sampling the function g at the frequency f_n Create x_s ^c _n , x^cp _n And g_n H is a composite function of_s ^c _n , h^cp _n Is calculated using fast Fourier transform (FFT).
[0020]
“Correlation function”
  In this apparatus, the sequence h synthesized by the synthesis function creation function 21_s ^c _n , h^cp _n (0 ≦ c ≦ C, 1 ≦ p ≦ N) is received, and the transmission side data sequence h_s ^c _n And one receiving side data series h^cp _n To obtain a new sequence r^cp _n Get. Here, C × N calculations are performed. If it is necessary for the quality degradation determination function 23, h_s ^c _n Is obtained by FFT, and a new sequence r is obtained._s ^c _n Ask for. If △ T is not required when probing traffic passively, h_s ^c _n Is calculated to obtain ΔT. Where r_s ^c _n , r^cp _n Against
[Outside 2]

The result of applying an exponential weight such as_s ^c _n , r^cp _n A method is also conceivable.
[0021]
"Quality degradation judgment function"
  In this apparatus, all data strings r are obtained from the cross-correlation calculation function 22._s ^c _n , r^cp _n (0.ltoreq.c.ltoreq.C, 1.ltoreq.p.ltoreq.N), and the range from 0 to .DELTA.T or the lag where the first peak appears is t^cpAnd t^cpTo t^cpExtract data in the range of + ΔT and replace it with new r_s ^c _n , r^cp _n And The number of data points is m = ΔT / f.
Moreover, quality evaluation is performed by one of the following methods using these.
・ R^cp _n Find the maximum and minimum values of^cpAnd
・ R^cp _n And find the maximum value of E^cpAnd
・
[Outside 3]

Seeking E to become
[Outside 4]

F^cpAsk for. Next, E^cpF^cpE divided by^cpIs normalized.
A series w in which a function that becomes a bathtub curve is sampled at a frequency f in a range from 0 to ΔT._m Prepare
[Outside 5]

Seeking E to become
[Outside 6]

F^cpAsk for. Next, E^cpF^cpE divided by^cpIs normalized.
[0022]
  Alternatively, E is obtained as follows without extracting the range of the data series.
r^cp _n And the spectral density of the frequency component having a period of ΔT is calculated as E^cpAnd R_s ^c _n The power spectrum is obtained, and the spectral density of the frequency component having a period of ΔT^cpIs normalized.
  Next, E^cp1 does not exceed a threshold value close to 1, 1^cp= 0, 1 if the threshold is exceeded^cp= 1. When monitoring a priority error, E^0P<E^1P<E^2P<... <E^NPMake sure that When this relationship is maintained, 1 for quality degradation determination result 1^P = 1. This indicates that there is no quality degradation between classes. If the above relationship is not maintained, 1^P = 0. This indicates that a priority error has occurred between classes. In the following, 1 in a specific class^cpAlso about 1^P Is written.
[0023]
Next, this operation is performed for all E.^P And E^cpOr h^cp _n Or r^cp _n Is the output value.
Continuous monitoring is realized by continuously performing the above operations.
[0024]
(Topology grasping device)
The topology grasping device 30 is realized by a route information estimation function 33, a route information collection function 34, a branch / junction / short-circuit removal function 31, and a logical topology creation function 32. The route information estimation function 33 and the route information collection function 34 can be realized by conventional techniques. For example, the path information estimation function 33 can be performed by observing a packet loss in the reception side packet rope, or when managing a packet propagation path, using the management information, or SNMP ( This can be realized by using a network management protocol such as Simple Network Management Protocol. Accordingly, the packet loss information is supplied from the packet probe device to the route information estimation function 33, and the route information estimated by the function 33 is supplied to the logical topology creation function 32. Here, the description will be continued assuming that the packet transfer path is obtained by these techniques. The packet transfer path is a path without a loop from one transmitter to the receiver.
[0025]
The route information collection function 34 collects route information by using the management information or network management protocol when managing the packet propagation route, and the collected route information is created by the branch / junction / short-circuit removal function 31 and logical topology creation. Supply to function 32.
[0026]
The branch / junction / short-circuit removal function 31 and the logical topology creation function 32 convert the packet transfer path obtained by the path information estimation function 33 or the path information collection function 34 into a logical topology that can be used by the quality degradation location determination device 40. Have The logical topology is a tree structure, and includes a node V, a link L, a root Vr, and a leaf Vl. Leaf Vl indicates a set of lines, routers, and switches, and usually indicates a branch point of a route including the routers and switches. A link L indicates a packet transfer path without a branch constituted by a line and a router. A route Vr indicates a transmission side packet probe, and a leaf Vl indicates a reception side packet probe. They are connected according to the following rules.
In Vr, one or more nodes V, or one or more leaves V1, or one or more nodes V and one or more leaves V1 are connected by a link L.
In the leaf V1, two or more nodes V, two or more leaves V1, or one or more nodes V and one or more leaves V1 are connected by a link L.
The leaf V1 is connected to one node V or one root Vr by a link L.
[0027]
The actual packet transfer route is set so that all routers and switches are unidirectionally connected when the route is a directed graph when viewed from a single packet transmitter. There is no strongly connected component. However, there is a possibility that the structure has a path that merges after branching. For this reason, in the present invention, in order to obtain a logical topology having a tree structure according to the above rules, the branch and merge short circuit removal function is found as a first step to find a path that branches and merges. All links and nodes in the closed area are short-circuited to obtain a tree-structured topology. When performing measurement using multicast, it is possible to directly output a multicast tree from the path information estimation function 33 or the path information collection function 34 to the logical topology creation function without using the branch / junction short-circuit removal function. It is. The presence of a strongly connected component (loop) is a problem of path control. In the present invention, it is assumed that the path control is normally performed and there is no strongly connected component.
[0028]
"Branch junction short circuit removal function"
In this apparatus, all the links in the area closed by branching and merging are short-circuit removed. As one method for performing short circuit removal, an embodiment of a branch / merge short circuit removal function by a method of color-coding a link in two colors will be described below.
A propagation token has one or more data tokens. The data token has data consisting of a set of the number of branches, branch node number, color, and initial leaf number. The data token has a data stack area composed of a set of the number of branches, a branch node number, a color, and a starting leaf number. Further, the nodes of the packet transfer path have node numbers that do not overlap. At this time, the token is flowed according to the following rules.
-On one link, only one propagation token can pass in the direction opposite to the direction of the directed tree (route table).
When the propagation token arrives at the node, all data tokens are extracted from the propagation token.
When output from a node, all data tokens are wrapped in a propagation token.
The initial attributes of the data token are the number of branches = none, the branch node number = none, the color = white, the stack is empty, and the leaf number where the token was generated is given as the initial leaf number. The initial attribute of the link is white.
-When a propagation token branches, that is, when a node with two or more inputs is reached, the data token is extracted from the propagation token, the current attributes are stored in the stack for all data tokens, and the number of branches = branch The number of links, color = red, branch node number = node number are given as new attributes. The first leaf number is not changed. Next, wrap all data tokens in one propagation token, create a copy of the propagation token, and stream the same propagation token on all links.
・ If there are multiple branches from the leaf, generate a data token with the same attributes as above, wrap the data token with one propagation token, create a copy of the propagation token, Run on all links.
-In the link where the propagation token including the data token with the red attribute flows, the link attribute is changed to red when the attribute is white. Link attributes never change from red to white.
When the tokens merge, that is, when a node having two or more outputs is reached, the following operations are performed after waiting for a propagation token to arrive from all links.
For data tokens with the same branch node number and the same starting leaf number and branch node number, if the number of data tokens is equal to the number of branches that is an attribute of the data token, the data token attribute is Swap the attributes from the stack into a single token. Repeat this process as much as possible.
When there is no data token to be collected, all data tokens are wrapped in one propagation token and output.
[0029]
By passing tokens from all leaves according to the above rules, the links can be color coded. A logical topology having a tree structure can be obtained by performing short circuit removal on the red portion in the color-coded links in this way to obtain one node.
[0030]
FIG. 3 shows an operation example. In FIG. 3, a thick solid line indicates a link having a red attribute, a thin solid line indicates a link having a white attribute, and an arrow direction indicates a direction indicated by a route table. Circles are nodes, and node numbers are shown in circles. A square is a data token, (a, b, c, d) indicates an attribute, and [(a, b, c, d)] indicates an attribute stored in the stack. Here, a indicates the number of branches, b indicates a branch node number, c indicates a color attribute, and d indicates a starting leaf number. In the stack, the attributes are stored so that they are packed up and retrieved sequentially from the top. A balloon indicates a propagation token.
[0031]
In FIG. 3, the token is illustrated as being issued from the starting point leaf number 1. Initially, the attributes of all links are white, and there is one data token in the propagation token, which has attributes (-,-, white, 1). This propagation token flows in the opposite direction on the directed graph. Since the node 6 is branched into two, the data token has the attribute (2, 6, red, 1), and the attribute (-,-, white, 1) is stored in the stack. Since a propagation token having a data token that is a red attribute has passed, the attribute of the link is red. Next, since the branch is performed again at the node 5, the data token has the attribute (2, 5, red, 1), and the attribute (2, 6, red, 1) is stored in the stack. Although the merge is performed in the node 2, since there is no data token that can be collected, all the data tokens are combined into one propagation token and passed upstream. At node 1, merging occurs. First, since two data tokens having the attribute (2, 5, red, 1) appear, they are collected as one data token and given the attribute (2, 6, red) extracted from the stack. Further, in node 1, since there are two data tokens having the attribute (2, 6, red, 1), these are also combined into one data token. As a result, as shown at the end of the figure, the merge and branch points are expressed as red links, and by removing this short circuit, the link can be changed to two links and one node.
[0032]
"Logical topology creation function"
In this apparatus, based on the tree structure obtained by the branching / merging / short-circuit removing function, a change is made so that the structure conforms to the rules described above. In the present apparatus, a portion composed of a node and a link having a single link input and a single link as an output is changed to a single link. FIG. 4 illustrates an actual change. In FIG. 4, circles indicate nodes, arrows indicate links, and squares indicate leaves. Further, the nodes that are short-circuited are indicated by hatching. A logical topology is created by the devices shown in the above (1) and (2), and becomes the output of the topology grasping device.
[0033]
(Deterioration device for quality degradation)
The quality degradation location determination device 40 is connected to the quality monitoring device 20 by the I^P This is a function for estimating a location where the quality has deteriorated by the degradation location determination function 41 and the degradation location estimation function 42. Hereinafter, description will be given in accordance with the processing order of these apparatuses.
`` Deterioration point judgment function ''
When the quality monitoring device 20 determines the occurrence of quality degradation, the degradation point determination function 41 receives the quality monitoring result from the quality monitoring device 20 and receives the logical topology from the topology grasping device 30. The degradation point determination function performs processing using these results.
[0034]
A basic determination method will be described with reference to FIG. The degradation point determination function is<sup>P</sup> Is received, and the lowered part is determined. When the topology information and the quality monitoring result shown on the left side of FIG. 5 are received, the route from the route Vr to the leaves Vl2 and Vl3 has little difference in transmission and reception patterns, that is, there is no packet loss and delay variation is small. It can be said that it is quality. However, quality degradation has occurred on the path to the leaf Vl1. Here, in the path to the leaf V11, quality degradation occurs at a place other than the path to the leaves 2 and 3, or at the output interface portion at the branch point of the path to the leaves 2 and 3 and the path to the leaf 1. it seems to do. Therefore, in the route to leaf 1, a link that is not included in the route to leaf 2 or 3 is marked, and further, a node that is a branch point of the route to leaf 1 from the route to leaf 2 or 3 is added. Put a mark. The part with this mark is output as a part where there is a high possibility that quality degradation has occurred. In FIG. 5, a link in which quality degradation has occurred is marked for clarity. As described above, the degradation point determination function 41 identifies the link portion where the quality degradation has occurred based on the quality degradation judgment result supplied from the quality monitoring device and the logical topology supplied from the topology grasping device 30.
[0035]
Hereinafter, detailed description will be given according to the structure of the logical topology. FIG. 6 illustrates topology information received from the topology grasping device, information on one class received from the quality monitoring device, or a priority error determination result. In FIG. 6, the numbers inside the leaves indicate the leaf numbers, and the contents enclosed in parentheses indicate the attributes of the leaves. (○) indicates that no quality degradation occurred, that is, I<sup>P</sup> = 1, and (x) indicates that quality degradation occurred, that is, I<sup>P</sup> = 0. Circles indicate nodes, and numbers inside the nodes indicate node numbers.
[0036]
First, in a node to which only the leaf is connected, attributes are assigned according to the following rules. In FIG. 6, nodes 7, 9, 10, 11, 13, 14, 15, 16, and 17 are targeted.
・ If all leaf attributes are (○). Assign (○, ○) attribute to the node.
・ If all leaf attributes are (×). Give the node the (x, x) attribute.
・ If the leaf attributes are (○) and (×). Give the node an attribute of (×, ○).
[0037]
This result is illustrated in FIG. Regarding the notation of (A, B), A indicates the quality state at the node, and this is called the own node state. B indicates the possibility that quality degradation has occurred in the upper part from the node, and this is called a propagation state.
[0038]
Next, attributes are assigned according to the following for nodes having attributes and nodes having only leaves as children. In the figure, 6, 8, and 12 are targeted.
When the propagation state of all child nodes is ○ and the leaf exists in the child, it is set as (◯, ○) when the attribute of all the child leaves is ○.
-When the propagation state of all child nodes is x and the leaf exists in the child, the attribute of all the child leaves is x. (×, ×).
・ If there is a leaf in the child, including the attribute of the leaf, if the propagation state of child nodes is mixed with ○ and ×, (×, ○).
[0039]
The attribute of the route Vr is the same as described above. The result of repeating the above operation is illustrated in FIG.
[0040]
When the own node state is “x”, it indicates that quality degradation can be observed in the lower node or leaf. There are two possibilities: a possibility that quality degradation has already occurred from the upper node and a possibility that it has occurred between the own node and the lower node including the own node. However, when the propagation state of the own node is ○, it indicates that there is a part where the quality degradation has not occurred in a part of the lower nodes, so the quality degradation occurs in the section from the upper node to the own node. It is considered that there is a low possibility. Therefore, it is considered that there is a high possibility that quality degradation has occurred in the link from the own node including the own node to the child node whose propagation state is x or the link to the leaf having the attribute (x).
[0041]
Based on the assumptions in the previous paragraph, a mark is assigned according to the following rules, and the set of switches and routers indicated by the marked nodes and links is determined as a place where there is a possibility that quality has deteriorated. The output of the function.
-A mark is given to a node whose own node state is x including the route.
-Mark a link from an individual node to which a mark is given to a child node whose propagation state is x or a link to a leaf having an attribute (x).
[0042]
The result is shown in FIG. When the attribute of the root is (x, x), a mark is given to a child node seen from the root or a leaf, and a link to that node or leaf is marked. In FIG. 9, it is impossible to determine in which section the quality degradation has occurred in the subtrees of the node numbers 1, 3, 6, 7, 15, and 14. Therefore, if there is a need to analyze such a partial tree, the degradation location estimation function 41 estimates a location where quality degradation may have occurred.
[0043]
`` Deterioration point estimation function ''
If a node or leaf having the (x, x) attribute constitutes a subtree, all the subtrees are extracted. Hereinafter, although an estimation method is shown for one of the extracted subtrees, the degradation point estimation function 42 performs the following processing for all the extracted subtrees. A data series E belonging to all the leaves in the subtree is used as a method for estimating the quality-decreasing portion.^CPHistorical data (time series data such as the last 24 hours 164 hours), h^CP _n Or r^CP _n A method of estimating the drop by obtaining the correlation using, can be considered. A brief description will be given with reference to FIG.
[0044]
In FIG. 10, it can be seen from the correlation between the upper and lower sides that the same packet time sequence has arrived at the leaves Vl2 and Vl3. However, packets arrive at the leaf Vl1 in a pattern different from the leaves Vl2 and 3. If there is no degradation point in the route after passing the degradation point, E^CP, H^CP _n , R^CP _n It is considered that the correlation between the data series increases. For this reason, in the upper part of the figure, it is considered that there is a possibility that quality degradation has occurred on the path from the root Vr to the leaf 1 and on the path from the root Vr to the node 2. In the lower part of the figure, it is considered that there is a possibility that quality degradation has occurred on the route from the root Vr to the leaf Vl1. In FIG. 10, a mark is given to the estimated location obtained in this way. The lowered location estimation function outputs a location with a mark as an estimated location. Hereinafter, a detailed description will be given using the tree structure shown in FIG.
[0045]
A subtree composed of leaves and nodes having only the leaves as children is made into one group. FIG. 12 shows the result of grouping. In FIG. 12, subtrees surrounded by a dotted line form a group. In each group, the data series E corresponding to the leaf^CPHistory data (time series data such as the past 24 hours, 164 hours), or h^CP _n Or r^CP _n Is used to find the strength of correlation for all combinations of leaves in the group. The result of the calculation. A combination with a high correlation is set to “◯”, and a combination with a low correlation is set to “X”. Using this result, attributes are assigned to the nodes and links in the group according to the following rules.
When all combinations in the group are “◯”, the attribute (O) is assigned to all nodes and links in the group.
-If all combinations in the group are "x", the (x) attribute is assigned to all nodes and links in the group.
-When "O" and "X" are mixed in the attribute of the combination, the attribute of (X) is given to the node, and further, two links leading to the two leaves corresponding to the combination of "O" The attribute (O) is assigned, and the attribute (X) is assigned to other links.
[0046]
The result is shown in FIG. As a method for reducing the number of correlation calculations, a method is conceivable in which a polygon having a leaf to be calculated at the vertex is created, a full mesh shape is drawn, and the correlation is obtained by paying attention to the triangle. For example, if there are five leaves consisting of a, b, c, d, and e, the vertices of the pentagon are a, b, c, d, and e, respectively, and the festival focuses on the triangles a, b, and c. When a−b is calculated as ◯ and bc is calculated as ×, a−c becomes ×. When cd is calculated as x, a−c needs to be calculated, and when a−e is calculated as ◯, b−e becomes ◯. These are summarized as follows.
[0047]
・ If the second side is o, the third side is o.
・ If one side is ○ and the other side is ×. The third side is marked with x.
・ If the two sides are x, calculate the third side.
[0048]
Next, a representative data series group is selected and given to the node as follows.
When all combinations in the group are “◯”, data corresponding to one leaf is arbitrarily set as a representative data series group.
When all combinations in the group are “x”, the data corresponding to all the leaves is set as the representative data series group.
・ When “◯” and “×” are mixed, pay attention to the combination that becomes “◯”, select the combination with the highest correlation from the related (closed) combinations, and select one arbitrary data Selected, others are excluded. For example, in FIG. 13, the closed combination in the group to which the node 8 belongs is leaves Vl5, 6, and 7, for example, the data of leaf Vl5 is included in the representative data series group. Further, in the node 10, since the two leaves Vl10 and 11, 12 and 22 are independently closed, for example, the data of the leaves 10 and 12 are included in the representative data series group. Data corresponding to all the leaves of the remaining combination “x” is also included in the representative data series group. The result is shown in FIG.
[0049]
Next, a node having an attribute assigned to a child and a leaf not subject to calculation in the previous stage and having no attribute assigned thereto is set as a processing target node. A subtree composed of these nodes and leaves is taken as a new group. Taking FIG. 14 as an example, nodes 4 and 11 are processing target nodes, and the group is surrounded by a dotted line.
[0050]
In each group, a correlation is obtained for a representative data series group possessed by a child node of the processing target node. At this time, the representative data series group is collected in units belonging to the child nodes as viewed from the processing target node, and this is referred to as a partial representative data series group. When there is a child leaf, the data series corresponding to the leaf is set as one partial representative data series group.
[0051]
The processing target node calculates cross-correlation between partial representative data series groups. No combination is made between data series in the partial representative data series group. For example, if the processing target node has child nodes A and B, the representative data series group of node A is 1 and 2, and the representative data series group of node B is 3 and 4, The data are grouped into two groups of partial representative data series, 3 and 4. The combinations are 1-3, 1-4, 2-3, 2-4.
[0052]
As a result of calculating the correlation between the partial representative data series groups, when there is even one combination that is “◯”, the correlation between the partial representative data series groups is (◯). When all the combinations are “x”, the correlation between the partial representative data series groups is (x).
Using the obtained correlation result, attributes are given to nodes and links as follows.
When the correlation between the partial representative data series groups is all (◯), the attribute “◯” is given to the processing target node and the link in the group.
When the correlation between the partial representative data series groups is all (×), the attribute “×” is given to the nodes and links in the group.
-When (○) and (×) are mixed in the correlation between partial representative data series groups, the link to the child node and the child node corresponding to the partial representative data series that is (○) An attribute is assigned, and an attribute “x” is given to the remaining child nodes and links. Further, an attribute “x” is given to the processing target node.
[0053]
Next, a representative data series group is selected. The result is shown in FIG.
When the combination in the group is “◯”, one data is arbitrarily selected from the combination having the highest correlation as the representative data series group.
When all combinations in the group are “x”, all data are set as representative data series groups.
・ If “○” and “×” are mixed in the combination, pay attention to the combination that is “◯” and select the combination with the highest correlation from the related (closed) combinations. Select one and exclude others. For example, in FIG. 15, there are two types of closed combinations 2-5 and 4-9 between the partial representative data series 7 and 8. In this case, for example, 2 and 4 are representative data, and 5 and 9 are excluded. Data that is not excluded from the remaining combinations of “x” is also set as the representative data series group.
[0054]
This result is illustrated in FIG. Thereafter, a new processing target node is determined and the same calculation is performed. The above operation is performed until there is no processing target node. Finally, the nodes and links having the attribute (X) are marked. An actual display example is shown in FIG.
[0055]
【The invention's effect】
According to the present invention, it is possible to monitor the state of quality control of a network that has performed priority control and grasp a control abnormality state such as a priority error. Further, when a quality abnormality occurs, it is possible to grasp the abnormal part. It is possible to eliminate the need for time-related quality measurement such as one-way delay time during quality measurement. As a result, the measurement devices dispersed in a plurality of locations realize only frequency synchronization, thereby enabling quality.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a quality state monitoring apparatus in a packet network.
FIG. 2 is an embodiment of a quality state monitoring apparatus in a packet network.
FIG. 3 is an example of an implementation method for removing a branching / merging structure from packet path information in a short time.
FIG. 4 is a modification example for creating a logical topology from a shortened topology.
FIG. 5 is an example of a basic grasping method for grasping a deteriorated part.
FIG. 6 is an example of output results of the quality monitoring device and the topology grasping device.
FIG. 7 is an example of a result of creating node attributes from leaf attributes;
FIG. 8 is an example of a result of creating a parent node attribute from a child node attribute;
FIG. 9 is an illustration of locations where it has been determined that quality degradation has occurred.
FIG. 10 is an illustration of a simple estimation method based on correlation.
FIG. 11 is an example of a directed tree that is processed by the drop location estimation function.
FIG. 12 is an example of grouping nodes and leaves.
FIG. 13 is an example of a grouped internal processing result.
FIG. 14 is an example of a selection result of a representative data series group.
FIG. 15 is an example of a processing result of a decrease location estimation function.
FIG. 16 is an example of a representative data series group selection result;
FIG. 17 is an example of a final quality degradation location estimation result.
[Explanation of symbols]
1 packet network
11, 12 Packet probe device
20 Quality monitoring device
21 Composite function creation function
22 Mutual function calculation function
23 Quality degradation judgment function
30 Topology grasp function
31 Short-circuit removal function
32 Logical topology creation function
33 Route information estimation function
34 Route information collection function
40 Degradation point judging device
41 Decline location function
50 Display device

Claims (19)

The network quality is monitored by comparing the transmission state and the reception state of the monitoring target packet, and when quality degradation is found, a method for determining a quality degradation occurrence section,
A step of calculating a cross-correlation function based on a packet passage time sequence obtained by probing a packet in each measurement device by installing a plurality of measurement devices in a network to be monitored for quality measurement ;
Determining a generation period of a packet or a packet group, extracting a time series in one period range, and performing a quality evaluation using the maximum correlation value in the time series;
If the quality degradation is determined to have occurred, create a logical topology as a tree structure from the pre-packet forwarding path information, that it comprises the step of determining quality degradation occurring point by utilizing the logical topology A quality state monitoring method in a packet network characterized by the above. Packets using a transmission side data sequence representing a packet transmission time sequence and a reception side data sequence representing a packet reception time sequence output from a plurality of packet probe devices arranged in a packet network to be monitored and having synchronized frequency sources A quality state monitoring method in a packet network that monitors the quality of a network and determines that a quality degradation portion has occurred when the quality degradation occurs in the packet network,
The transmission side data series and the reception side data series are received, a cross correlation function is obtained using the transmission side data series and the reception side data series, and it is determined whether or not quality degradation has occurred from the obtained cross correlation function. Outputting a quality evaluation result of the packet network;
Receiving packet transfer route information of the packet network and creating a logical topology having a tree structure from the received packet transfer route information;
A quality state monitoring method in a packet network, comprising the step of determining a quality degradation part based on the output quality evaluation result and the created logical topology .   3. The quality status monitoring method for a packet network according to claim 1 or 2, wherein a test packet is generated to dynamically monitor the network quality.   The method of using a multicast as a test packet generation method in the quality status monitoring method in the packet network according to claim 3.   4. A method of using a plurality of unicasts as a test packet generation method in the quality status monitoring method in a packet network according to claim 3.   3. A quality state monitoring method in a packet network according to claim 1, wherein after the cross-correlation function is obtained, the quality evaluation is performed using the maximum correlation value. 3. The quality state monitoring method for a packet network according to claim 2 , wherein after the cross-correlation function is obtained, a generation period of the packet or the packet group is determined, a time series in one period range is extracted, and a correlation value in the time series is obtained. Quality evaluation using the maximum value of.   3. The quality state monitoring method for a packet network according to claim 1 or 2, wherein after the cross-correlation function is obtained, a generation period of the packet or the packet group is determined, a time series in one period range is extracted, and the time series within the time series is determined. A method of obtaining the maximum and minimum correlation values and evaluating the quality using these differences.   The quality state monitoring method for a packet network according to claim 1 or 2, wherein an autocorrelation function of a packet transmission time sequence is obtained after obtaining a cross-correlation function, a generation period of the packet or the packet group is determined, Take the time series, find the integral value of the cross-correlation function and the auto-correlation function within the time series, and evaluate the quality using the value obtained by dividing the integral value of the cross-correlation function by the integral value of the auto-correlation function Method.   The quality state monitoring method for a packet network according to claim 1 or 2, wherein the generation period of a packet or a group of packets is determined after obtaining a cross-correlation function, and a time series within one period is extracted, and the extracted time series In this method, exponential weighting is performed by the vertical axis method, the maximum value and the minimum value of the correlation value within the time series are obtained, and quality evaluation is performed using these differences.   3. A quality state monitoring method in a packet network according to claim 1 or 2, wherein a cross-correlation function is obtained, then an auto-correlation function of a packet transmission time series is obtained, a generation period of the packet or the packet group is determined, and one period range The time series is extracted, and the extracted time series is exponentially weighted by the vertical axis method to obtain the integral value of the cross-correlation function and the auto-correlation function within the time series, and the integration of the cross-correlation function A method of evaluating quality using the value divided by the integrated value of the autocorrelation function.   3. The quality status monitoring method for a packet network according to claim 1 or 2, wherein when the occurrence of the quality degradation is determined, a closed region such as branching / merging is short-circuited by propagating a token with attributes on a logical topology. A method of obtaining a tree structure by removing.   3. The quality state monitoring method for a packet network according to claim 1 or 2, wherein a location where quality degradation has occurred is determined from end / end measurement results, and quality degradation is determined when a quality degradation location is observed from the measurement results. A method of obtaining a link and a node indicating a quality degradation by removing nodes and links that are considered not to have a quality degradation using a logical topology structure from the obtained path.   14. A quality state monitoring method in a packet network according to claim 13, wherein when a degraded part is determined as a subtree, a correlation between end and end measurement results is obtained between leaves, and the result and the structure of the subtree are used. This is a method for estimating the quality degradation point more finely.   15. The quality monitoring method for a packet network according to claim 1, wherein the packet network is classified into transfer ranks with packet transfer priorities as transfer classes, and priority transfer control is performed for each class. A method characterized in that the network is configured to perform QoS control.   16. A network monitoring method, comprising: monitoring a priority error in a preferentially forwarded network using the quality state monitoring method in a packet network according to claim 1. Packets using a transmission side data sequence representing a packet transmission time sequence and a reception side data sequence representing a packet reception time sequence output from a plurality of packet probe devices arranged in a packet network to be monitored and having synchronized frequency sources A quality monitoring device for a packet network that monitors the quality of the network and, when it is found that the quality has deteriorated in the packet network,
The transmission side data series and the reception side data series are received, a cross correlation function is obtained using the transmission side data series and the reception side data series, and it is determined whether or not quality degradation has occurred from the obtained cross correlation function. A quality monitoring device for outputting the quality evaluation result of the packet network;
A topology grasping device that receives packet transfer route information of the packet network and creates a logical topology having a tree structure from the received packet transfer route information;
A quality state of a packet network, comprising: a quality degradation location determination device for judging a quality degradation location based on a quality evaluation result output from the quality monitoring device and a logical topology output from a topology grasping device Monitoring device.   18. The quality state monitoring apparatus for a packet network according to claim 17, wherein the quality monitoring apparatus generates a synthesis function on the transmission side and a synthesis function on the reception side for each of the transmission side data series and the reception side data series. Means, a cross-correlation calculating means for calculating a cross-correlation between the transmitting-side combining function and the receiving-side combining function, and a quality deterioration determining means for determining whether or not a quality deterioration has occurred from the obtained cross-correlation A device characterized by comprising:   The quality monitoring apparatus for a packet network according to claim 17 or 18, wherein the topology grasping device receives packet loss information from the packet probe device and estimates route information of the packet; Route information collecting means for collecting packet route information from the management information, branch / junction / short circuit removing means for short-circuiting branching and / or merging from the route information output from the route information collecting means, and the route information estimating means An apparatus comprising: a logical topology creating means for creating a logical topology based on the output and the output from the branch merging / short circuit removing means.

Images