研究課題/領域番号 |
21KK0204
|
研究種目 |
国際共同研究加速基金(国際共同研究強化(A))
|
配分区分 | 基金 |
審査区分 |
小区分60010:情報学基礎論関連
|
研究機関 | 北陸先端科学技術大学院大学 |
研究代表者 |
シュワルツマン グレゴリー 北陸先端科学技術大学院大学, 先端科学技術研究科, 准教授 (20815261)
|
研究期間 (年度) |
2022 – 2023
|
研究課題ステータス |
交付 (2022年度)
|
配分額 *注記 |
10,530千円 (直接経費: 8,100千円、間接経費: 2,430千円)
|
キーワード | Theory / Networking / Graph algorithms |
研究開始時の研究の概要 |
We aim to achieve the following two goals: 1) Develop a theoretical framework for P4 switches 2) Show that theoretical tools can be applied in practice to solve real-world problems in P4 switches We will achieve this via close collaborations between project members, both theoreticians and practitioners
|
研究実績の概要 |
During my stay in Europe we achieved great progress in the project. On the practical side, we implemented many theory primitives (spanners, synchronizers, spanning trees) on P4 switches and checked their efficiency at handling failures in datacenter networks. The main problems we considered are clock synchronization and multicast. We show that using our techniques we can recover from failures considerably faster than other techniques.
On the theory side we defined a theoretical model for distributed computation with bounded memory. We believe that this two way approach will greatly increase the communication between the theory and networking community.
|
現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Currently we have preliminary results in submission both at networking and theory conferences. As we are introducing new ideas to both communities, it might take some time until the results get accepted. In parallel, we are working on polishing existing results and introducing new results from both the theory and networking perspectives.
|
今後の研究の推進方策 |
This year I plan to visit the UK again to polish up our theoretical model. The main focus will be to consider the connection between streaming algorithms and limited memory distributed computation. I will also attend the ICLR conference, with the hope of combining machine learning techniques and our networking results.
|