אירועים והרצאות בפקולטה למדעי המחשב ע"ש הנרי ומרילין טאוב

Nowadays, as computer networks continuously increase in size and speed, making efficient use of compute and network resources becomes exceptionally challenging. At the heart of this challenge are scheduling and load balancing techniques that are expected to optimize resource usage and, in turn, minimize costs. In this talk we shall address two such emerging network challenges.

In the first part of the talk we shall present dRMT (disaggregated Reconfigurable Match-Action Table), a new architecture for programmable switches. dRMT overcomes two significant restrictions of RMT, the predominant pipeline-based architecture for programmable switches, namely: (i) table memory is local to an RMT pipeline stage, implying that memory not used by one stage cannot be reclaimed by another, and (ii) RMT is hard-wired to always sequentially execute matches followed by actions as packets traverse pipeline stages. dRMT resolves both issues by disaggregating the memory and compute resources of a programmable switch. At the heart of dRMT is a new scheduling technique that can execute programs at line rate with fewer processors compared to RMT, and it avoids performance cliffs when there are not enough processors to run a program at line rate.

In the second part of the talk we shall focus on load balancing. Specifically, we will introduce “Local-Shortest-Queue” (LSQ) which is a new family of load balancing algorithms. Roughly speaking, in LSQ each load balancer (a.k.a. dispatcher) keeps a local view of the server queue lengths and routes jobs to the shortest among them. Communication is used only to update the local views and make sure that they are not too far from the real queue lengths. We will show how LSQ policies significantly outperform well-known low-communication policies that have been considered for the multi-dispatcher scenario, such as JSQ(d) and JIQ, while using a similar communication budget.

* Ph.D. Advisors: Prof. Ariel Orda and Prof. Isaac Keslassy.