JP2004236236A - Buffering apparatus and method for asynchronously reaching variable-length packet - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a buffering apparatus for asynchronously reaching variable-length packets, in which a packet discard rate can be reduced by performing priority control with the variable-length packets, and to provide a buffering method. <P>SOLUTION: The buffering apparatus is provided with a queue composed of a plurality of delay elements for temporarily storing the packets of transmitted digital signals, a first optical switch for distributing the packets to the plurality of delay elements, a second optical switch for selecting for outputting the packets that pass through the plurality of delay elements, a reading means for reading the priority and length of the packet, and a buffer management device for controlling the first optical switch and the second optical switch according to the read priority and length. When discarding a packet located between the reading means and the first optical switch, the first optical switch is used to discard the packet, but when discarding a packet located between the first optical switch and the second optical switch, the second optical switch is used to discard the packet. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

【００２１】
ここで、関数Ｃｅｉｌ（ｘ）は、ｘの少数点以下を切り上げた整数をあらわす。Δ_ｋ＜Ｂであれば、そのパケットをΔ_ｋ番目の遅延線に格納する（Ｆ２）。また、Δ_ｋ≧Ｂであれば、そのパケットを棄却する。このように一部は棄却し、他は格納するパケットであるので、それに応じてバッファ占有量ｆを更新する必要が生じる。そこで、パケットを格納する時には（Ｆ４）、
【００２２】
【数２】

【００２３】
とｆを更新する。棄却されるポートを除いて、全ポートのパケットの遅延を求めた後、さらにｍａｘ（ｆ−Ｔ、０）を新たなバッファ占有量ｆとして更新し（Ｆ１１）、Ｔ時間後の次のラウンドロビン周期のパケット処理に備える。
【００２４】
この様に光遅延線バッファを用いると、ポートｋに到着したパケットは、その直前に格納されたパケット、すなわち、ポートｋに到着したパケットの遅延をもとめる際に用いられるｆで示された時間に出力されるパケットとは、次の間隔をもつ。
【数３】

【００２５】
この間隔が大きい場合は、バッファに無駄な空きが生じるので、棄却率が増加することが予想される。一方、Ｄを小さくして間隔を詰める場合には、バッファを有効利用できるが、バッファの長さＢ×Ｄが小さくなり格納できるパケットの数が少なくなるので、棄却率が増加することが予想される。したがって、遅延線の単位長Ｄの大きさが、通信の品質を定める重要なパラメータとなる事がわかる。
【００２６】
また、可変長パケットで優先制御を行なうバッファ管理法については、ＰＢＳを拡張した方法が非特許文献７にＷＡ／ＴＤとして提案されている。このＷＡ／ＴＤは、ＰＢＳを可変長非同期に適用したものであるが、ポート間の処理順や、周期を区切ることは明記されていない。以下では、このバッファ管理法を用いた方式をｖＰＢＳ方式と称する。
【００２７】
ｖＰＢＳ方式ではＰＢＳ方式と同じく、閾値より後方には優先度の低いパケットを格納することは許さない。このため、優先度の高いパケット棄却率を優先度の低いパケット棄却率より小さくすることができるという利点はあるが、優先制御を行なわないバッファ制御であるため、本発明に比べてパケット棄却率が大きくなる。
【００２８】
【非特許文献１】
葉原 敬士、三条 広明、西沢 秀樹、小川 育夫、須崎 泰正、“波長ルーチング型大容量光パケットスイッチの開発”、 電子情報通信学会技術研究報告（ＳＳＥ９９−１４８）， ２０００年２月．
【非特許文献２】
Ｄａｖｉｄ Ｋ． Ｈｕｎｔｅｒ ａｎｄ Ｍｅｏｗ Ｃ． Ｃｈｉａ ａｎｄ Ｉｖａｎ Ａｎｄｏｎｏｖｉｃ， “Ｂｕｆｆｅｒｉｎｇ ｉｎ ＯｐｔｉｃａｌＰａｃｋｅｔ Ｓｗｉｔｃｈｅｓ”， ＩＥＥＥ／ＯＳＡ Ｊｏｕｒｎａｌ ｏｆ Ｌｉｇｈｔｗａｖｅ Ｔｅｃｈｎｏｌｏｇｙ， Ｖｏｌ． １６， Ｎｏ． １２，ｐｐ． ２０８１−２０９４， Ｄｅｃｅｍｂｅｒ １９９８．
【非特許文献３】
Ｙ． Ｌｉｎ ａｎｄ Ｊ． Ａ． Ｓｉｌｖｅｒｓｔｅｒ， “Ｐｒｉｏｒｉｔｙ Ｑｕｅｕｉｎｇ Ｓｔｒａｔｅｇｉｅｓ ａｎｄＢｕｆｆｅｒＡｌｌｏｃａｔｉｏｎ Ｐｒｏｔｏｃｏｌｓ ｆｏｒ Ｔｒａｆｆｉｃ Ｃｏｎｔｒｏｌ ａｔ ａｎ ＡＴＭ Ｉｎｔｅｇｒａｔｅｄ ＢｒｏａｄｂａｎｄＳｗｉｔｃｈｉｎｇ Ｓｙｓｔｅｍ”， ＩＥＥＥ Ｊｏｕｒｎａｌ ｏｎ Ｓｅｌｅｃｔｅｄ Ａｒｅａｓ ｉｎ Ｃｏｍｍｕｎｉｃａｔｉｏｎｓ， Ｖｏｌ． ９，Ｎｏ． ９， ｐｐ． １５２４−１５３６， Ｄｅｃｅｍｂｅｒ １９９１
【非特許文献４】
Ｌ． Ｋｌｅｉｎｒｏｃｋ， Ｑｕｅｕｅｉｎｇ ｓｙｓｔｅｍｓ Ｖｏｌｕｍｅ ＩＩ： Ｃｏｍｐｕｔｅｒ ａｐｐｌｉｃａｔｉｏｎｓ． ＡＷｉｌｅｙ−Ｉｎｔｅｒｓｃｉｅｎｃｅ Ｐｕｂｌｉｃａｔｉｏｎ， １９７６．
【非特許文献５】
Ｓ． Ｌ． Ｄａｎｉｅｌｓｅｎ ａｎｄ Ｂ． Ｍｉｋｋｅｌｓｅｎ ａｎｄ Ｃ． Ｊｏｅｒｇｅｎｓｅｎ ａｎｄ Ｔ． Ｄｕｒｈｕｕｓ ａｎｄ Ｋ．Ｅ． Ｓｔｕｂｋｊａｅｒ， “ＷＤＭ Ｐａｃｋｅｔ Ｓｗｉｔｃｈ Ａｒｃｈｉｔｅｃｔｕｒｅｓ ａｎｄ Ａｎａｌｙｓｉｓ ｏｆ ｔｈｅ Ｉｎｆｌｕｅｎｃｅ ｏｆＴｕｎａｂｌｅ Ｗａｖｅｌｅｎｇｔｈ Ｃｏｎｖｅｒｔｅｒｓ ｏｎ ｔｈｅ Ｐｅｒｆｏｒｍａｎｃｅ”， ＩＥＥＥ／ＯＳＡ Ｊｏｕｒｎａｌ ｏｆＬｉｇｈｔｗａｖｅ Ｔｅｃｈｎｏｌｏｇｙ， Ｖｏｌ． １５， Ｎｏ． ２， ｐｐ．２１９−２２７， Ｆｅｂｒｕａｒｙ １９９７．
【非特許文献６】
Ｆ． Ｃａｌｌｅｇａｔｉ， “Ｏｐｔｉｃａｌ ｂｕｆｆｅｒｓ ｆｏｒ ｖａｒｉａｂｌｅ ｌｅｎｇｔｈ ｐａｃｋｅｔｓ，”ＩＥＥＥＣｏｍｍｕｎｉｃａｔｉｏｎｓ Ｌｅｔｔｅｒｓ， ｖｏｌ． ４， ｐｐ． ２９２−２９４， Ｓｅｐｔｅｍｂｅｒ ２０００．
【非特許文献７】
Ｆ． Ｃａｌｌｅｇａｔｉ， Ｇ． Ｃｏｒａｚｚａ， ａｎｄ Ｃ． Ｒａｆｆａｅｌｌｉ， “Ｅｘｐｌｏｉｔａｔｉｏｎ ｏｆ ＤＷＤＭ ｆｏｒｏｐｔｉｃａｌ ｐａｃｋｅｔ ｓｗｉｔｃｈｉｎｇ ｗｉｔｈ ｑｕａｌｉｔｙ ｏｆ ｓｅｒｖｉｃｅ ｇｕａｒａｎｔｅｅｓ，” ＩＥＥＥ Ｊｏｕｒｎａｌ ｏｎＳｅｌｅｃｔｅｄ Ａｒｅａｓ ｉｎ Ｃｏｍｍｕｎｉｃａｔｉｏｎｓ， ｖｏｌ． ２０， ｐｐ． １９０−２０１， Ｊａｎｕａｒｙ ２００２．
【非特許文献８】
Ｎ． Ｗａｄａ， Ｈ． Ｈａｒａｉ， Ｗ． Ｃｈｕｊｏ， ａｎｄ Ｆ． Ｋｕｂｏｔａ， “Ｐｈｏｔｏｎｉｃ ｐａｃｋｅｔ ｒｏｕｔｉｎｇ ｂａｓｅｄｏｎ ｕｌｔｉｗａｖｅｌｅｎｇｔｈ ｌａｂｅｌ ｓｗｉｔｃｈｉｎｇ ｕｓｉｎｇ ｆｉｂｅｒ Ｂｒａｇｇ ｇｒａｔｉｎｇｓ，” ＥＣＯＣ ２０００Ｔｅｃｈｎｉｃａｌ Ｄｉｇｅｓｔ （２６ｔｈ Ｅｕｒｏｐｅａｎ Ｃｏｎｆｅｒｅｎｃｅ ｏｎ Ｏｐｔｉｃａｌ Ｃｏｍｍｕｎｉｃａｔｉｏｎ）， ｖｏｌ． ４（Ｎｏ． １０．４．６）， ｐｐ． ７１−７２， Ｓｅｐｔｅｍｂｅｒ ２０００．
【００２９】
【発明が解決しようとする課題】
上記のように、可変長パケットで優先制御を行なうバッファ管理法については、既に報告された例があるが、棄却率を低減する余地がまだ残っている。
【００３０】
この発明は上記に鑑み提案されたもので、可変長パケットで優先制御を行なってパケットの棄却率を低減することのできる、非同期到着する可変長パケットのバッファリング装置とその方法を提供することを目的とする。
【００３１】
【課題を解決するための手段】
上記目的を達成するために、第１の発明は、非同期到着する可変長パケットのバッファリング装置に関しており、伝送されたデジタル信号のパケットを一時的に保管する複数の遅延素子からなるキューと、上記のパケットを上記の複数の遅延素子に振り分ける第１の光スイッチと、上記の複数の遅延素子を通過したパケットを選択して出力に振り向ける第２の光スイッチと、パケットの優先度と長さとを読取る読取り手段と、その読み出された優先度と長さとにより、第１の光スイッチと第２の光スイッチとを制御するバッファ管理装置と、を、備え、上記の読取り手段と第１の光スイッチとの間に有るパケットを棄却する場合は第１の光スイッチを用いて棄却し、第１の光スイッチと第２の光スイッチとの間に有るパケットを棄却する場合は第２の光スイッチを用いて棄却することを特徴としている。
【００３２】
また、第２の発明は、伝送されたデジタル信号のパケットを一時的に保管する複数の遅延素子からなるキューと、上記のパケットを上記の複数の遅延素子に振り分ける第１の光スイッチと、パケットの優先度と長さとを読取る読取り手段と、その読み出された優先度と長さとにより、第１の光スイッチを制御するバッファ管理装置とを備える非同期到着する可変長パケットのバッファリング装置に摘要するバッファリング方法であって、
上記のキューには予め決められた閾値が設けられており、上記のキューに新たなパケットを追加する際に、上記のキューに格納されたパケットの並びについて、
１）上記のキューに格納されたパケットのバッファ占有量が上記の閾値を越えなければ、優先度に関わらず、上記のキューに格納されたパケットの並びの最後方にその追加するパケットを格納し、
２）上記のキューに格納されたパケットのバッファ占有量が上記の閾値をこえる場合であって、
２Ａ）その追加するパケットが、上記のキューに格納されたパケットのどれよりも優先度の低いパケットの場合は、上記のキューに格納されたパケットの並びの最後方にその追加するパケットを格納し、
２Ｂ）その追加するパケットに比べて優先度が同等か高いパケットが、上記のキューに格納されたパケットの中にある場合には、その優先度が同等か高いパケットの次の後方にその追加するパケットを格納して、その格納する位置に、より優先度の低いパケットがあれば、そのパケットを棄却して格納し、
２Ｃ）その追加するパケットが、上記のキューに格納されたパケットのどれよりも優先度の高いパケットの場合は、上記の閾値を越えて格納されているパケットの並びの最後方にその追加するパケットを格納するが、その格納する位置に、より優先度の低いパケットがあれば、そのパケットを棄却して格納することを特徴としている。
【００３３】
【発明の実施の形態】
以下にこの発明の実施の形態を図面に基づいて詳細に説明する。以下においては、同等の機能を有する装置等については同じ符号を用いるものとする。まず、可変長パケットで優先制御を行なうバッファ管理法について説明する。
【００３４】
説明を分かり易くするために、ひとつの例として、優先度が高い優先度１のクラスと優先度が低い優先度２のクラスとバッファへ格納するパケット数についての閾値があるものとする。さらに多くの優先度を想定した場合については、下記と同様の様式で容易に拡張することができる。このバッファ管理法では、次のようにする。
【００３５】
１）まず、すでにバッファに格納されているパケットによるバッファ占有量が閾値を越えていなければ、優先度に関わらず到着したパケットをキューの最後方に格納する。
【００３６】
２）一方、占有量が閾値をこえる場合には、クラス２のパケットはキューの最後方に格納される。クラス１のパケットは、閾値を越えて格納されているクラス１のパケットの最後方に格納される。その位置にクラス２のパケットがあれば、そのパケットは棄却される。棄却されたパケットは上書きされたと見なされる。
【００３７】
これをより一般化すると次の様になる。まず、すでに上記のキューに格納されたパケットの並びがあり、
キューには予め決められた複数の閾値が設けられており、上記のキューには予め決められた複数の閾値が設けられており、それらの閾値について、昇順に並べられたそれぞれの閾値は、昇順あるいは降順にならべられた単数あるいは連続する複数の優先度に対応付けられているものとし、それぞれの閾値に属する優先度は、そのそれぞれの単数あるいは連続する複数の優先度の最も優先する優先度とするとき、上記のキューに新たなパケット（Ｐａ）を追加する際に、
０）上記のキューに格納された最後方のパケットが上記の複数の閾値のいずれにも届かない場合には、優先度に関わらずパケットの並びの最後方にその追加するパケット（Ｐａ）を格納し、
１）上記のキューに格納された最後方のパケット（Ｐｒ）が上記の複数の閾値のいずれか（Ｔｈａ）により分断される様に配置しているか、その最後方のパケット（Ｐｒ）の最後尾が上記の閾値（Ｔｈａ）に一致している場合に、優先度に関わらず、上記のキューに格納されたパケットの並びの最後方にその追加するパケットを格納し、
２）上記のキューに格納されたパケットのバッファ占有量が前記閾値（Ｔｈａ）をこえていて、最後方のパケットが前記閾値（Ｔｈａ）を分断される様に配置されていない場合であって、
２Ａ）その追加するパケット（Ｐａ）が、上記のキューに格納されたパケットのどれよりも優先度の低いパケットの場合は、上記のキューに格納されたパケットの並びの最後方にその追加するパケット（Ｐａ）を格納し、
２Ｂ）その追加するパケット（Ｐａ）が上記閾値（Ｔｈａ）に対応づけられた優先度以上に優先されるパケットであって、上記のキューの上記閾値より後方に、上記閾値（Ｔｈａ）に対応づけられた優先度以上に優先されるパケット（Ｐｂ）が格納されたパケットの中にある場合には、そのパケット（Ｐｂ）の次の後方にその追加するパケット（Ｐａ）を格納するが、その格納する位置に、上記閾値（Ｔｈａ）に対応付けられた優先度以下で優先されないパケット（Ｐｃ）があれば、そのパケット（Ｐｃ）を棄却して追加するパケット（Ｐａ）を格納し、
２Ｃ）その追加するパケット（Ｐａ）が、上記閾値（Ｔｈａ）に対応づけられた優先度以上に優先されるパケットであって、上記のキューの上記閾値より後方に上記閾値（Ｔｈａ）に対応づけられた優先度以上に優先されるパケットが、格納されたパケットの中にない場合には、上記閾値（Ｔｈａ）で分断されるパケットの後方にその追加するパケットを格納するが、その格納する位置に、上記閾値（Ｔｈａ）に対応付けられた優先度以下で優先されないパケット（Ｐｃ）があれば、そのパケットを棄却して格納し、
２Ｄ）その追加するパケット（Ｐａ）が、上記閾値（Ｔｈａ）に対応づけられた優先度以下で優先されないパケットであれば、上記のキューに格納されたパケットの並びの最後方にその追加するパケット（Ｐａ）を格納する。
【００３８】
このバッファ管理法を用いたバッファ管理装置を図６に示し、そのパケットに対する振舞いを、図７、８、９のフローチャートに従って次に説明する。図７、８、９は、横に連続する図であり、優先度が２クラスの場合を示している。さらに優先度クラスを増加させる場合は、図７に相当する処理を付け加えれば良い。また、図７のＭ２で示した部分を下記のアルゴリズムにしたがって変更する。
【００３９】
図６（ａ）は、それに用いるＮ入力×１出力の光バッファ（Ｎ＝４）を示している。バッファリング装置は、Ｂ個のファイバ遅延線４、Ｎ×（Ｂ＋１）個の光スイッチ１、Ｂ×２個の光スイッチ２、バッファ管理装置３、より構成される。到着するパケットＡ、Ｂ、Ｃ、Ｄの優先度を、それぞれ、２、２、１、２とし、バッファ管理装置の閾値をＴｈ＝Ｄとしたときのバッファ管理装置の振舞いを例として示す。図６（ｂ）は、パケットＡ（優先度＝２）が遅延線ｄ_０に切替えられた直後の出力ポートから見たパケットの位置を示している。この場合は、パケットＡ（優先度＝２）を光バッファに格納した後で、バッファ占有度は閾値Ｔｈを超えていることがわかる。したがって、パケットＢ（優先度＝２）については、上記の２）の方針に従って、一旦バッファに格納されるが、この直後に処理されるクラス１のパケットＣによって上書される。ここでいう上書き機能は、まだ入っていないパケットは前段の光スイッチで棄却し、すでに遅延線を流れているパケットは後段の光スイッチで棄却することで実現する。クラス２のパケットＤは、パケットＣの後方に格納される。
【００４０】
次に、このバッファ管理法を用いたバッファ管理装置で持つ情報を以下に述べる。
【００４１】
優先度の種類と等しいｐ個の閾値（Ｔｈ_１、Ｔｈ_２、・・・、Ｔｈ_ｐ）を設ける。それぞれは、遅延線の単位長Ｄの倍数である。
また、遅延線の数をＢとして、
【００４２】
【数４】

とする。
【００４３】
優先度１、２、・・・、ｐのパケットについて、優先度ごとに、該当優先度のパケットがバッファへの格納を許される最小の時刻 ｆ_１、ｆ_２、・・・、ｆ_ｐのそれぞれを、上記の可変長パケットであるが優先制御を行なわないバッファ管理法の場合に習って定義する。ここで、バッファ占有量ｂ＝ｆ_ｐとする。
【００４４】
このとき以下のアルゴリズムに従って処理をするものとする。
（１） ｎ＝１、２、・・・、Ｎに対して、ポートｎにパケットが到着していれば、以下の処理を順に行なう。
（１−１） そのパケットは優先度ｉ、パケット長ｌ_ｎｉ で、周期開始からｔ_ｎ遅れて到着したとする。光遅延線バッファの離散特性により、パケットは、
【００４５】
【数５】

【００４６】
の遅延を持つ必要がある。Δ_ｎ＜Ｂなら、そのパケットをΔ_ｎ番目の遅延線に格納する（図７のＦ３）。また、Δ_ｎ≧Ｂであれば、そのパケットを棄却（図７のＦ９９）する。
【００４７】
（１−２） パケットをバッファに格納する時は、次のポートに到着するパケットを適切に処理するために、優先度ｉをもつパケットに関する変数ｆ_ｉについて、
【００４８】
【数６】

【００４９】
とする（図７のＦ４）。さらに、他の変数（または、該当優先度のパケットがバッファへの格納を許される最小の時刻）ｆ_１、ｆ_２、・・・、ｆ_ｐも以下のように変更する。
【００５０】
Ａ）到着したパケットを格納する前のバッファ占有量ｂが、Ｔｈ_ｐより小さければ、到着するパケットの優先クラスに関わらず、ｆ_ｉの値をｆ_ｋに設定する（ｋ＝１、２、・・・、ｐ）
【００５１】
Ｂ） 到着したパケットを格納する前のバッファ占有量ｂが、
Ｔｈ_ｊ≦ｂ＜Ｔｈ_ｊ−１ （１＜ｊ≦ｐ）であり、
優先度ｊ以下のパケット（ｉ≧ｊ）が到着する場合、
ｋ＝ｊ、ｊ＋１、・・・、ｐ に対してｆ_ｉの値をｆ_ｋに設定する。
【００５２】
Ｃ） 到着したパケットを格納する前のバッファ占有量ｂが、
Ｔｈ_ｊ≦ｂ＜Ｔｈ_ｊ−１ （１＜ｊ≦ｐ） であり、
クラスｊより優先されるパケット（ｉ＜ｊ） が到着する場合、
ｋ＝１、２、・・・、ｊ−１ に対して、
ｆ_ｉの値をｆ_ｋに設定する。優先度ｍがｊまたはそれより低いパケット（ｍ≧ｊ）を格納する位置を、クラスがｊより高いパケットより前方に格納しないように、ｍ＝ｊ、ｊ＋１、・・・、ｐに対して、
ｆ_ｍにｍａｘ（ｆ_ｊ−１、ｆ_ｍ）の値を設定する。
【００５３】
（２）また、全てのポートのパケットが処理された後、次の周期の処理を適切に行なうために、ｆ_１、ｆ_２、・・・、ｆ_ｐ を変更する。
【００５４】
Ｄ） バッファ占有量ｂ が閾値Ｔｈ_ｐ以下であれば、
ｋ＝１、２、・・・、ｐに対して、ｆ_ｋにｍａｘ（ｆ_ｋ−Ｔ、０）を設定する。
【００５５】
Ｅ） バッファ占有量ｂ が閾値Ｔｈ_ｉ≦ｂ＜Ｔｈ_ｉ−１ （１＜ｉ≦ｐ）であれば、ｍ＝ｉ、ｉ＋１、・・・、ｐに対して、ｆ_ｍにｆ_ｍ−Ｔを設定する。
また、ｋ＝１、２、・・・、ｉ−１に対して、
ｆ_ｋにｍｉｎ（ｆ_ｐ、ｍａｘ（ｆ_ｋ−Ｔ、Ｔｈ_ｉ））の値を設定する。
【００５６】
上記の処理を高速で行う装置は、図１０、１１、１２に示したフローチャートをもとにゲートアレイを構成することにより容易に実現することができる。図１０、１１、１２、１３は一連の図面であり、符号は、図７、８、９において対応するブロックを示している。
【００５７】
このバッファリング方法を用いると、パケットの処理順序は、バッファへの到着順序とは異なる。しかし、優先度に注目すると、それぞれのポートにおけるパケットの到着順序は、棄却されたパケットを除外すると、その出発順序と同じである。このように、出発と到着の順序が保たれていることは、他のアプリケーションにも好都合であることはあきらかである。
【００５８】
このバッファリング方法についてのシミュレーション結果を図１３に示す。これは、１０^９個のパケットを用いてシミュレーションした結果である。ここで、パケットは、率ａのポアソン過程に従って到着するものとし、優先度１と優先度２とのパケット数の比率は、ρ１：ρ２とした。パケットの長さは、平均１で、指数関数的に分布するものとした。遅延線の数を５０とし、その遅延単位時間をＤ＝１．０とした。Ｄは平均パケット長で正規化している。ここで、閾値の値を従来の方法（ｖＰＢＳ）に比べ本発明の方法（ｖＰＢＳＯ）を僅かに小さくすることによってパケットの棄却率の改善が図れることが明らかになっている。これは、これらの方法の本質的な違いに依るものである。図１３から分かるように、本発明の方法（ｖＰＢＳＯ）では、従来の方法（ｖＰＢＳ）に比べて、パケットの棄却率が小さくなっている。
【００５９】
【発明の効果】
非同期到着する可変長パケットについて優先制御を行なうようにしたので、優先するパケットの棄却率を優先されないパケットの棄却率よりも低減することができた。また、優先度の低いパケットを上書きする本発明（ｖＰＢＳＯ）により、従来のＰＢＳ方式を可変長パケットに拡張するｖＰＢＳ方式よりも、パケット棄却率を低減することができた。
【図面の簡単な説明】
【図１】一般的に用いられるバッファの状態を表すキューを示す図である。
【図２】従来の光パケットスイッチアーキテクチャの例を示す図である。
【図３】部分バッファ共有方式（ＰＢＳ）に用いられるキューを示す図である。
【図４】４入力×１出力の光バッファ構成を示す図である。
【図５】従来の光バッファ構成を示す図である。
【図６】本発明の光バッファ構成を示す図である。
【図７】本発明のバッファリング方法を実行するためのフローチャートを示す図である。
【図８】本発明のバッファリング方法を実行するためのフローチャートを示す図である。
【図９】本発明のバッファリング方法を実行するためのフローチャートを示す図である。
【図１０】本発明のバッファリング装置に用いるゲートアレイを構成する際に用いるフローチャートを示す図である。
【図１１】本発明のバッファリング装置に用いるゲートアレイを構成する際に用いるフローチャートを示す図である。
【図１２】本発明のバッファリング装置に用いるゲートアレイを構成する際に用いるフローチャートを示す図である。
【図１３】本発明のバッファリング方法についてのシミュレーション結果を示す図である。
【符号の説明】
１、２ 光スイッチ
３ バッファ管理装置
４ 光遅延素子
５ 優先度と長さの読取り手段[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for buffering variable-length packets arriving asynchronously, which realizes optical communication using packets of various lengths using an optical circuit having a limited buffer capacity and holding time.
[0002]
[Prior art]
With the development of optical communication technology capable of efficiently transmitting a large amount of information, optical technology is being introduced not only in transmission paths but also in packet switching in order to realize higher throughput. Generally, in packet switching, a packet transfer process is performed, and a packet switch is used for this. The functions of the packet switch for this purpose can be roughly classified into five types: (1) destination search, (2) exchange, (3) buffer management, (4) buffering, and (5) path control. Generally, a packet switch in which the function of the above (2) or (4) is made optical is called an optical packet switch. It has a wide band characteristic by being composed of an optical switch and an optical buffer and transferring data as it is as an optical signal. Furthermore, recently, the above-mentioned (1) destination search can be performed at high speed by optical analog processing such as multi-wavelength label processing and optical code label processing. When these label processes are combined with an optical switch, an ultra-high performance 1 input × N (natural number) output optical packet switch can be constructed. Hereinafter, a packet switch that performs the above functions (1), (2) and (4) as it is with light is referred to as an optical packet switch.
[0003]
Further, by improving the above (3) buffer management, the transfer capability of the optical packet switch can be further improved. Conventionally, a logic circuit using an electronic circuit has been used for (3) buffer management. This is because there is no optical logic circuit at a practical level.
[0004]
One of the purposes of buffer management is to prevent packet collisions. In a packet switching network, when a packet arrives at a packet switch at the same time, the packet is stored in a buffer to avoid collision. FIG. 1 is a diagram showing a queue indicating a state of a buffer generally used. For example, in the current Internet, packet transfer without priority control is performed as shown in FIG. Packets arriving at the packet switch are stored in the buffer, if available, regardless of priority.
[0005]
In addition, a semiconductor memory is often used for the buffer. However, Non-Patent Document 1 or Non-Patent Document 2 reports an optical buffer control mounting method having a buffer, even though the optical packet switch does not use a semiconductor memory.
[0006]
A plurality of optical fibers having different lengths are used in an already known optical buffer. These are optical delay elements having different delay times, and are stored during the delay time. It is known that collision between packets can be avoided by controlling the packets allocated to the optical delay line. At this time, since it is not possible to take out the packet at an arbitrary time, it is necessary to prevent the collision of the packet at the time of output. Therefore, when putting the packet into the optical buffer, it is necessary to manage the delay time using some buffer management technique. .
[0007]
FIG. 2 shows, as an example of a conventional optical packet switch architecture, a configuration including N 1 input × N output bufferless packet switches of an N input × N output optical packet switch and N N input × 1 output buffers. Is shown. Each 1 input × N output switch and N input × 1 output buffer are connected in a substantially grid pattern. In order to configure an N-input × N-output packet switch that fully demonstrates the performance of the 1-input × N-output packet switch, it is necessary to process a plurality of packets arriving at the same time in a sufficiently short time. Therefore, it is important to apply a buffer management method with a small amount of calculation.
[0008]
As described in Non-Patent Document 8, the 1-input × N-output bufferless packet switch enables high-speed destination search by an optical destination search function. As a result, the N-input × N-output optical packet switch exhibits a high node throughput. However, a buffer is required to avoid packet collision and improve the rejection rate. An N input × 1 output buffer is connected to each output port of the N input × N output packet switch.
[0009]
Also, recently, with the emergence of various applications, it is desired to provide priority control by a network. This is a method in which a packet is prioritized and transmitted. If the packet has a low priority, it may be rejected. Buffer management methods for realizing such priority packet transfer include a partial buffer sharing method (PBS; Partial Buffer Sharing), a push-out method (PO; Push Out), and the like, which are described in Non-Patent Document 3. ing.
[0010]
In the partial buffer sharing method (PBS), as shown in FIG. 3, when the number of packets in the buffer exceeds a threshold value, storage of a newly arriving non-priority packet in the buffer is not allowed. The number of packets in the buffer is managed by one variable (queue length), and the value is changed according to the arrival of the packet and the passage of time. In the PBS, a threshold value is predetermined for the queue length, that is, the sum (or the number of packets) of the packet lengths entering the queue. If the queue length is less than the threshold, all packets will enter the queue. On the other hand, if the queue length is equal to or greater than the threshold, only priority packets (hereinafter, packets at discard level 1) are stored in the queue, and non-priority packets (discard level 2) are discarded. The amount of calculation required to change the variable is small, and packet processing can be performed in a relatively short time. In this method, the overall packet loss rate is considerably higher than the transfer by the method without priority transfer control. However, it is possible to make the rejection rate of a high-priority packet lower than that of a low-priority packet.
[0011]
In the push-out method (PO), when a high-priority packet arrives when the buffer is full, a low-priority packet already in the buffer is expelled and stored. As a result, it is possible to make the rejection rate of a high-priority packet lower than that of a low-priority packet as in the case of the PBS. In addition, the overall packet loss rate is smaller than that of the PBS, and the same performance as when priority control is not performed can be achieved. However, in the PO method, it is necessary to manage not only the number of packets but also where the non-priority packets are located, and the management of the packets in the buffer is complicated. In this management, the location information of a non-priority packet is used not only to determine which packet is to be discarded, but also to the position where the discarded packet was located, and to the priority packet located after that to the front. , Is also used to store the last priority packet that arrives.
[0012]
In a router or an ATM switch having a conventional electronic buffer, a buffer management method is known in which the buffer is divided into a plurality of parts and each is managed as one queue. The multiple queue method using this buffer management method is an effective method for priority control because the queues are individually managed and the packets taken out simultaneously by the buffer management device are limited to those from one queue.
[0013]
On the other hand, it is difficult to realize the multiple queue system by using multiple queues using an optical fiber delay line buffer. This is because each optical fiber delay line buffer allows only a specific delay time.
[0014]
As a method using a single queue, the partial buffer sharing method (PBS) described above is already known. As another priority control method using a single queue, Non-Patent Document 4 describes HOL (Head-of-the-Line) priority queue control, and Non-Patent Document 3 describes a push-out method (PO: Push Out). It is described in. In the above-mentioned HOL priority queue control method or push-out method (PO), all of the buffers are shared regardless of the priority of the packet, so that the use efficiency of the optical buffer increases in those methods. However, in the HOL system, when a packet of the discard level 1 arrives, the non-priority packets in the queue are shifted one by one. In the PO method, when a priority packet arrives when the buffer is full, the last non-priority packet in the queue is discarded, and priority packets behind it move one by one. Therefore, assuming that the maximum queue length is B, in the above-described HOL priority queue control method or push-out method (PO), the calculation amount according to the algorithm becomes a value dependent on B in any case. The relatively large amount of calculation in this way causes a decrease in buffer management speed. When using an optical packet switch in an area with a high transmission speed, the simple PBS method is more effective than the HOL or PO method. Often it is.
[0015]
Next, a buffer management method that is a variable-length packet but does not perform priority control will be described more specifically. This method is already well known. For example, it can be managed by the following method in which round robin scheduling is added to Non-Patent Document 6.
[0016]
The output buffer type N input × N output packet switch shown in FIG. 2 uses an N input × 1 output optical buffer and N 1 × 1 × N output switches. When the transmission amount increases, there is a possibility that a maximum of N packets arrive at the same time in the optical buffer of N inputs × 1 output.
[0017]
In order to prevent collision (superposition) of these packets, a delay line buffer is used. In this case, in order to store all the packets in the buffer, the processing for obtaining the delay of N packets is performed by the time l corresponding to the minimum packet length. _min Must be done within. That is, the allowable time for processing one packet is l _min / N. This time is equivalent to the processing time of one packet when buffering is performed according to a simple round robin. This indicates that buffer management can be performed by round robin scheduling.
[0018]
FIG. 4A shows an optical buffer configuration of 4 inputs × 1 output. In this case, N = 4. The optical buffer composed of optical delay elements is composed of B optical delay elements using fiber delay lines, N × (B + 1) optical switches, and a buffer management device. B fiber delay lines ｛d ₀ , D ₁ , ..., d _B-1 Has a length of a multiple {0, D,..., (B−1) × D} of the unit length D of the delay line. In addition, the round robin scheduling is repeated every T time by a signal from the buffer management device. For this reason, the buffer management device receives the information of the packet at a point of time T or more away from the preceding optical switch, and uses this information during the predetermined time T to use each of the N packets. The required delay time is calculated. During the time T, it is necessary to process continuously arriving packets, so that T ≦ l _min Must be satisfied. For example, as shown in FIG. 4A, four packets arrive and packets A, B, and C are respectively delayed by the buffer management device by delay lines d. ₀ , D ₂ , D ₃ And the packet D is rejected. Such control is realized by the buffer management device driving the preceding optical switch. FIG. 4B shows that the packet A has a delay line d. ₀ Shows the position of the packet as seen from the output port immediately after switching. As shown in FIG. 4C, packets A, B, and C are output without collision. In this example, the packet B arrives at the buffer before the packet A, but the packets that arrive are sequentially processed in a round-robin manner, so that the packets are output in the order shown in FIG.
[0019]
Hereinafter, the behavior of the buffer management device will be specifically described with reference to the flowchart of FIG. The buffer management device performs a process of avoiding a packet collision at the time of output or inside. Here, a predetermined time t during the round robin repetition time T ₀ From the time t when all packets in the optical buffer leave the buffer _f Is f. Hereinafter, f is referred to as a buffer occupancy. The buffer management device obtains a delay time of a packet arriving at all ports for each cycle T. The delay time given to packets arriving within the same cycle is obtained in the order of port k = 1, 2,..., N. Packet length l on port k _k Packet at time t ₀ With the origin as time t _k Suppose you arrive at The delay time t to be given to a packet for collision avoidance is t = ft _k It is. Here, considering that the delay line buffer has discrete time characteristics, it is understood that the packet should be delayed by the next time. (F1 in FIG. 5)
[0020]
(Equation 1)

[0021]
Here, the function Ceil (x) represents an integer obtained by rounding up the decimal point of x. Δ _k If <B, the packet is Δ _k The data is stored in the second delay line (F2). Also, Δ _k If ≧ B, the packet is rejected. As described above, some packets are discarded and others are packets to be stored. Therefore, it is necessary to update the buffer occupancy f accordingly. Therefore, when storing the packet (F4),
[0022]
(Equation 2)

[0023]
And f are updated. After calculating the delays of the packets of all the ports except for the ports to be discarded, max (fT, 0) is further updated as a new buffer occupancy f (F11), and the next round robin after T time Prepare for periodic packet processing.
[0024]
When the optical delay line buffer is used in this way, the packet arriving at the port k can be stored in the packet stored immediately before it, that is, the time indicated by f used to determine the delay of the packet arriving at the port k. The output packet has the following intervals.
[Equation 3]

[0025]
If this interval is large, a wasteful space is generated in the buffer, and the rejection rate is expected to increase. On the other hand, when the interval is reduced by making D smaller, the buffer can be used effectively, but the buffer length B × D becomes smaller and the number of packets that can be stored becomes smaller, so that the rejection rate is expected to increase. You. Therefore, it is understood that the size of the unit length D of the delay line is an important parameter that determines the quality of communication.
[0026]
As a buffer management method for performing priority control with variable-length packets, a method that extends the PBS is proposed in Non-Patent Document 7 as WA / TD. In this WA / TD, a PBS is applied in a variable length asynchronous manner, but it is not specified that a processing order between ports or a period is delimited. Hereinafter, a system using this buffer management method is referred to as a vPBS system.
[0027]
In the vPBS method, similarly to the PBS method, storing a packet having a lower priority behind a threshold value is not allowed. Therefore, there is an advantage that the high-priority packet drop rate can be made smaller than the low-priority packet drop rate. However, since the buffer control does not perform the priority control, the packet drop rate is lower than that of the present invention. growing.
[0028]
[Non-patent document 1]
Takashi Habara, Hiroaki Sanjo, Hideki Nishizawa, Ikuo Ogawa, Yasumasa Suzaki, "Development of Wavelength Routing Type Large Capacity Optical Packet Switch", IEICE Technical Report (SSE99-148), February 2000.
[Non-patent document 2]
David K. Hunter and Meow C.E. Chia and Ivan Andonovic, "Buffering in Optical Packet Switches", IEEE / OSA Journal of Lightwave Technology, Vol. 16, No. 12, pp. 2081-2094, December 1998.
[Non-Patent Document 3]
Y. Lin and J.M. A. Silverster, "Priority Queuing Strategies and BufferAllocation Protocols for Traffic Control at an ATM Integrated Broadcasting Joint Economics System. 9, No. 9, pp. 1524-1536, December 1991
[Non-patent document 4]
L. Kleinrock, Queueing systems Volume II: Computer applications. AWiley-Interscience Publication, 1976.
[Non-Patent Document 5]
S. L. Danielsen and B.A. Mikkelsen and C.M. Joergensen and T.W. Durhuus and K. E. FIG. J. Stubkjaer, "WDM Packet Switch Architecture and Analysis of the Influence of Tunable Wavelength Converters on the Performance of the Internet of Things." 15, No. 2, pp. 219-227, February 1997.
[Non-Patent Document 6]
F. Callegati, "Optical buffers for variable length packets," IEEE Communications Letters, vol. 4, pp. 292-294, September 2000.
[Non-Patent Document 7]
F. Callegati, G .; Corraza, and C.E. Raffaelli, "Exploitation of DWDM formal packet switching with quality of service guarantees," IEEE Journal on Selected Areas in Essentials Insurance. 20 pp. 190-201, January 2002.
[Non-Patent Document 8]
N. Wada, H .; Harai, W.M. Chujo, and F.S. Kubota, "Photonic packet routing basedon multiwavelength label switching using fiber fiber gratings," ECOC 2000 Technical Digest, 26th Annual Medical Digest (26th Economic Cog.). 4 (No. 10.4.6), pp. 2-5. 71-72, September 2000.
[0029]
[Problems to be solved by the invention]
As described above, a buffer management method of performing priority control using variable-length packets has already been reported, but there is still room for reducing the rejection rate.
[0030]
The present invention has been proposed in view of the above, and it is an object of the present invention to provide a buffering apparatus for a variable-length packet arriving asynchronously and a method thereof, which can perform priority control on the variable-length packet and reduce the packet rejection rate. Aim.
[0031]
[Means for Solving the Problems]
In order to achieve the above object, a first invention relates to a buffering device for a variable-length packet that arrives asynchronously, and a queue comprising a plurality of delay elements for temporarily storing a packet of a transmitted digital signal; A first optical switch for distributing the packet to the plurality of delay elements, a second optical switch for selecting a packet passing through the plurality of delay elements and directing the packet to an output, and the priority and length of the packet. Reading means, and a buffer management device for controlling the first optical switch and the second optical switch based on the read priority and length, and the reading means and the first When rejecting a packet between an optical switch and a first optical switch, rejecting a packet between a first optical switch and a second optical switch It is characterized in that reject using the second optical switch.
[0032]
According to a second aspect of the present invention, there is provided a queue including a plurality of delay elements for temporarily storing transmitted digital signal packets, a first optical switch for distributing the packets to the plurality of delay elements, and a packet. The buffering device for asynchronously arriving variable-length packets, comprising: reading means for reading the priority and length of the packet; and a buffer management device for controlling the first optical switch based on the read priority and length. Buffering method,
The queue is provided with a predetermined threshold, and when a new packet is added to the queue, the arrangement of the packets stored in the queue is
1) If the buffer occupancy of the packet stored in the queue does not exceed the threshold, the packet to be added is stored at the end of the queue of the packet stored in the queue regardless of the priority. ,
2) When the buffer occupancy of the packet stored in the queue exceeds the threshold,
2A) If the packet to be added has a lower priority than any of the packets stored in the queue, the packet to be added is stored at the end of the sequence of the packets stored in the queue. ,
2B) If a packet having the same or higher priority than the packet to be added is among the packets stored in the queue, the packet is added to the rear of the packet next to the packet with the same or higher priority. Store the packet, and if there is a lower priority packet at the storage location, discard the packet and store it.
2C) If the packet to be added has a higher priority than any of the packets stored in the queue, the packet to be added is located at the end of the sequence of packets stored exceeding the threshold. Is stored. If there is a lower priority packet at the storage location, the packet is rejected and stored.
[0033]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following, the same reference numerals are used for devices having the same functions. First, a description will be given of a buffer management method for performing priority control with variable-length packets.
[0034]
In order to make the description easy to understand, it is assumed that, as an example, there are a class having a high priority and a class having a low priority and a threshold regarding the number of packets to be stored in the buffer. Assuming more priorities can be easily extended in the same manner as described below. In this buffer management method, the following is performed.
[0035]
1) First, if the buffer occupancy of the packets already stored in the buffer does not exceed the threshold value, the arriving packet is stored at the end of the queue regardless of the priority.
[0036]
2) On the other hand, if the occupancy exceeds the threshold, the class 2 packet is stored at the end of the queue. The class 1 packet is stored at the end of the class 1 packet stored exceeding the threshold. If there is a class 2 packet at that position, the packet is discarded. Dropped packets are considered overwritten.
[0037]
This can be generalized as follows. First, there is a sequence of packets already stored in the above queue,
The queue is provided with a plurality of predetermined thresholds, and the queue is provided with a plurality of predetermined thresholds, and the thresholds are arranged in ascending order in the ascending order. Alternatively, it is assumed that they are associated with one or more consecutive priorities arranged in descending order, and the priority belonging to each threshold is the highest priority of the respective one or more plural priorities. When adding a new packet (Pa) to the above queue,
0) If the last packet stored in the queue does not reach any of the plurality of thresholds, the packet (Pa) to be added is stored at the end of the packet sequence regardless of the priority. And
1) Whether the last packet (Pr) stored in the queue is arranged so as to be divided by any of the plurality of thresholds (Tha), or the last packet (Pr) of the last packet (Pr) Is equal to the threshold value (Tha), the packet to be added is stored at the end of the sequence of the packets stored in the queue, regardless of the priority.
2) A case where the buffer occupancy of the packet stored in the queue exceeds the threshold (Tha) and the last packet is not arranged so as to be separated from the threshold (Tha),
2A) If the packet (Pa) to be added has a lower priority than any of the packets stored in the queue, the packet to be added is located at the end of the sequence of the packets stored in the queue. (Pa), and
2B) The packet to be added (Pa) is a packet that has a higher priority than the priority associated with the threshold value (Tha), and is associated with the threshold value (Tha) behind the threshold value of the queue. If the packet (Pb) having a higher priority than the given priority is included in the stored packet, the packet (Pa) to be added is stored after the packet (Pb). If there is a non-priority packet (Pc) below the priority associated with the threshold value (Tha) at the position where the packet (Pc) is dropped, the packet (Pc) is discarded and the packet (Pa) to be added is stored.
2C) The packet to be added (Pa) is a packet that has a higher priority than the priority associated with the threshold (Tha), and is associated with the threshold (Tha) behind the threshold of the queue. If there is no packet with higher priority than the given priority among the stored packets, the additional packet is stored behind the packet divided by the threshold value (Tha). If there is a non-priority packet (Pc) below the priority associated with the threshold (Tha), the packet is discarded and stored,
2D) If the packet (Pa) to be added is a packet that is not prioritized below the priority associated with the threshold (Tha), the packet to be added is located at the end of the list of packets stored in the queue. (Pa) is stored.
[0038]
FIG. 6 shows a buffer management device using this buffer management method. The behavior of the packet will be described below with reference to the flowcharts of FIGS. FIGS. 7, 8, and 9 are diagrams that are continuous horizontally and show a case where the priority is two classes. To further increase the priority class, a process corresponding to FIG. 7 may be added. Also, the portion indicated by M2 in FIG. 7 is changed according to the following algorithm.
[0039]
FIG. 6A shows an N-input × 1-output optical buffer (N = 4) used therein. The buffering device includes B fiber delay lines 4, N × (B + 1) optical switches 1, B × 2 optical switches 2, and a buffer management device 3. The behavior of the buffer management device when the priorities of the arriving packets A, B, C, and D are 2, 2, 1, and 2, and the threshold value of the buffer management device is Th = D is shown as an example. FIG. 6B shows that the packet A (priority = 2) has a delay line d ₀ Shows the position of the packet as viewed from the output port immediately after the switching. In this case, after storing the packet A (priority = 2) in the optical buffer, it can be seen that the buffer occupancy exceeds the threshold Th. Therefore, the packet B (priority = 2) is temporarily stored in the buffer according to the policy of 2) above, but is overwritten by the class 1 packet C processed immediately thereafter. The overwriting function here is realized by rejecting a packet that has not yet entered by the optical switch at the preceding stage, and rejecting a packet already flowing through the delay line by the optical switch at the subsequent stage. The packet D of class 2 is stored after the packet C.
[0040]
Next, information held in a buffer management device using this buffer management method will be described below.
[0041]
P thresholds equal to the type of priority (Th ₁ , Th ₂ , ..., Th _p ) Is provided. Each is a multiple of the unit length D of the delay line.
Also, assuming that the number of delay lines is B,
[0042]
(Equation 4)

And
[0043]
.., P, the minimum time at which the packet of the priority is allowed to be stored in the buffer for each priority f ₁ , F ₂ , ..., f _p Are defined in the case of the above-described buffer management method which is a variable-length packet but does not perform priority control. Here, the buffer occupancy b = f _p And
[0044]
At this time, processing is performed according to the following algorithm.
(1) If a packet arrives at port n for n = 1, 2,..., N, the following processing is performed in order.
(1-1) The packet has priority i and packet length l _ni And from the start of the cycle t _n Suppose you arrive late. Due to the discrete characteristics of the optical delay line buffer,
[0045]
(Equation 5)

[0046]
Need to have a delay. Δ _n If <B, the packet is Δ _n The data is stored in the third delay line (F3 in FIG. 7). Also, Δ _n If ≧ B, the packet is rejected (F99 in FIG. 7).
[0047]
(1-2) When storing a packet in a buffer, in order to appropriately process a packet arriving at the next port, a variable f relating to a packet having a priority i is used. _i about,
[0048]
(Equation 6)

[0049]
(F4 in FIG. 7). Further, another variable (or the minimum time at which the packet of the corresponding priority is allowed to be stored in the buffer) f ₁ , F ₂ , ..., f _p Is also changed as follows.
[0050]
A) The buffer occupation amount b before storing the arrived packet is Th _p If it is smaller than f, regardless of the priority class of the arriving packet. _i The value of f _k (K = 1, 2,..., P)
[0051]
B) The buffer occupancy b before storing the arrived packet is
Th _j ≦ b <Th _j-1 (1 <j ≦ p),
When a packet with priority j or lower (i ≧ j) arrives,
k = j, j + 1,..., p _i The value of f _k Set to.
[0052]
C) The buffer occupancy b before storing the arrived packet is
Th _j ≦ b <Th _j-1 (1 <j ≦ p), and
When a packet (i <j) that has priority over class j arrives,
For k = 1, 2,..., j−1,
f _i The value of f _k Set to. The position for storing a packet with a priority m of j or lower (m ≧ j) is not stored in front of a packet whose class is higher than j. For m = j, j + 1,.
f _m Max (f _j-1 , F _m ) Value.
[0053]
(2) Also, after packets of all ports have been processed, f ₁ , F ₂ , ..., f _p To change.
[0054]
D) Buffer occupancy b is equal to threshold Th _p If
For k = 1, 2,..., p, f _k Max (f _k −T, 0) is set.
[0055]
E) Buffer occupancy b is threshold Th _i ≦ b <Th _i-1 If (1 <i ≦ p), then for m = i, i + 1,. _m To f _m Set -T.
Also, for k = 1, 2,..., I−1,
f _k To min (f _p , Max (f _k -T, Th _i )).
[0056]
An apparatus that performs the above processing at high speed can be easily realized by configuring a gate array based on the flowcharts shown in FIGS. FIGS. 10, 11, 12, and 13 are a series of drawings, and reference numerals indicate corresponding blocks in FIGS.
[0057]
With this buffering method, the order of processing packets is different from the order of arrival in the buffer. However, paying attention to the priority, the order of arrival of packets at each port is the same as the order of departure except for discarded packets. It is clear that maintaining the order of departure and arrival is also advantageous for other applications.
[0058]
FIG. 13 shows a simulation result of this buffering method. This is 10 ⁹ It is a result of simulation using the number of packets. Here, it is assumed that the packets arrive according to the Poisson process at the rate a, and the ratio of the number of packets of the priority 1 and the priority 2 is ρ1: ρ2. The length of the packet was 1 on average and distributed exponentially. The number of delay lines was 50, and the delay unit time was D = 1.0. D is normalized by the average packet length. Here, it is clear that the packet rejection rate can be improved by slightly reducing the threshold value in the method of the present invention (vPBSO) as compared with the conventional method (vPBS). This is due to the essential differences between these methods. As can be seen from FIG. 13, the method of the present invention (vPBSO) has a smaller packet loss rate than the conventional method (vPBS).
[0059]
【The invention's effect】
Since the priority control is performed on the variable-length packets that arrive asynchronously, the rejection rate of the priority packets can be reduced than the rejection rate of the non-priority packets. In addition, the present invention (vPBO) that overwrites low-priority packets can reduce the packet loss rate compared to the vPBS method that extends the conventional PBS method to variable-length packets.
[Brief description of the drawings]
FIG. 1 is a diagram showing a queue representing a state of a buffer generally used.
FIG. 2 is a diagram illustrating an example of a conventional optical packet switch architecture.
FIG. 3 is a diagram showing queues used in a partial buffer sharing system (PBS).
FIG. 4 is a diagram showing a 4-input × 1-output optical buffer configuration.
FIG. 5 is a diagram showing a conventional optical buffer configuration.
FIG. 6 is a diagram showing an optical buffer configuration of the present invention.
FIG. 7 is a diagram showing a flowchart for executing the buffering method of the present invention.
FIG. 8 is a diagram showing a flowchart for executing the buffering method of the present invention.
FIG. 9 is a diagram showing a flowchart for executing the buffering method of the present invention.
FIG. 10 is a diagram showing a flowchart used when configuring a gate array used in the buffering device of the present invention.
FIG. 11 is a diagram showing a flowchart used when configuring a gate array used in the buffering device of the present invention.
FIG. 12 is a diagram showing a flowchart used when configuring a gate array used in the buffering device of the present invention.
FIG. 13 is a diagram showing a simulation result of the buffering method of the present invention.
[Explanation of symbols]
1,2 optical switch
3 Buffer management device
4 Optical delay element
5 Priority and length reading means

Claims (5)

A queue consisting of a plurality of delay elements for temporarily storing transmitted digital signal packets,
A first optical switch for distributing the packet to the plurality of delay elements;
A second optical switch for selecting a packet that has passed through the plurality of delay elements and redirecting the packet to an output;
Reading means for reading the priority and length of the packet;
A buffer management device that controls the first optical switch and the second optical switch based on the read priority and length;
With,
When rejecting a packet existing between the reading means and the first optical switch, the packet is rejected using the first optical switch, and a packet existing between the first optical switch and the second optical switch is discarded. A buffering device for asynchronously arriving variable-length packets, wherein the buffering is performed using a second optical switch when the packet is rejected. A queue consisting of a plurality of delay elements for temporarily storing transmitted digital signal packets,
A first optical switch for distributing the packet to the plurality of delay elements;
Reading means for reading the priority and length of the packet;
A buffering method for a buffering device for asynchronously arriving variable-length packets, comprising a buffer management device for controlling a first optical switch, based on the read priority and length, wherein A predetermined threshold is provided, and when adding a new packet to the queue, the arrangement of packets stored in the queue is
1) If the buffer occupancy of the packet stored in the queue does not exceed the threshold, the packet to be added is stored at the end of the queue of the packet stored in the queue regardless of the priority. ,
2) When the buffer occupancy of the packet stored in the queue exceeds the threshold,
2A) If the packet to be added has a lower priority than any of the packets stored in the queue, the packet to be added is stored at the end of the sequence of the packets stored in the queue. ,
2B) If a packet having the same or higher priority than the packet to be added is among the packets stored in the queue, the packet is added to the rear of the packet next to the packet with the same or higher priority. Store the packet, and if there is a lower priority packet at the storage location, discard the packet and store it.
2C) If the packet to be added is a packet having a higher priority than any of the packets stored in the queue, the packet to be added is located at the end of the list of packets stored exceeding the threshold. A buffering method for asynchronously arriving variable-length packets, characterized in that if a lower-priority packet is stored at the storage location, the packet is discarded and stored. A queue consisting of a plurality of delay elements for temporarily storing transmitted digital signal packets,
A first optical switch for distributing the packet to the plurality of delay elements;
Reading means for reading the priority and length of the packet;
A buffering method for a buffering device for asynchronously arriving variable-length packets, comprising a buffer management device for controlling a first optical switch, based on the read priority and length, wherein It is assumed that a plurality of determined thresholds are provided, and for each of the thresholds, the thresholds arranged in ascending order are associated with one or more consecutive priorities arranged in ascending or descending order. , When the priority belonging to each threshold is the highest priority of each single or a plurality of consecutive priorities,
When adding a new packet (Pa) to the above queue,
0) If the last packet stored in the queue does not reach any of the plurality of thresholds, the packet (Pa) to be added is stored at the end of the packet sequence regardless of the priority. And
1) Whether the last packet (Pr) stored in the queue is arranged so as to be divided by any of the plurality of thresholds (Tha), or the last packet (Pr) of the last packet (Pr) Is equal to the threshold value (Tha), the packet to be added is stored at the end of the sequence of the packets stored in the queue, regardless of the priority.
2) A case where the buffer occupancy of the packet stored in the queue exceeds the threshold (Tha) and the last packet is not arranged so as to be separated from the threshold (Tha),
2A) If the packet (Pa) to be added has a lower priority than any of the packets stored in the queue, the packet to be added is located at the end of the sequence of the packets stored in the queue. (Pa), and
2B) The packet to be added (Pa) is a packet that has a higher priority than the priority associated with the threshold value (Tha), and is associated with the threshold value (Tha) behind the threshold value of the queue. If the packet (Pb) having a higher priority than the given priority is included in the stored packet, the packet (Pa) to be added is stored after the packet (Pb). If there is a non-priority packet (Pc) below the priority associated with the threshold value (Tha) at the position where the packet (Pc) is dropped, the packet (Pc) is discarded and the packet (Pa) to be added is stored.
2C) The packet to be added (Pa) is a packet that has a higher priority than the priority associated with the threshold (Tha), and is associated with the threshold (Tha) behind the threshold of the queue. If there is no packet with higher priority than the given priority among the stored packets, the additional packet is stored behind the packet divided by the threshold value (Tha). If there is a non-priority packet (Pc) below the priority associated with the threshold (Tha), the packet is discarded and stored,
2D) If the packet to be added (Pa) is a packet that is not given priority below the priority associated with the threshold value (Tha) and is not given priority, the packet to be added is located at the end of the list of packets stored in the queue. (Pa). A buffering method for variable-length packets arriving asynchronously, characterized by storing (Pa). The buffering apparatus for variable-length packets arriving asynchronously according to claim 1, which operates by the buffering method for variable-length packets arriving asynchronously according to claim 2. 3. The buffering apparatus for asynchronously arriving variable-length packets according to claim 1, wherein the apparatus operates according to the buffering method for asynchronously arriving variable-length packets.

Images