CS CAP Seminar: Jovan Stojkovic, "Memory-Efficient Hashed Page Tables” and “SpecFaaS: Accelerating Serverless Applications with Speculative Function Execution"

Event Type

Seminar/Symposium

Sponsor

Saugata Ghose

Location

Siebel 2405

Date

Nov 15, 2022   4:00 pm  

Views

42

Originating Calendar

Computer Science Speakers Calendar

We look forward to seeing you in person on Tuesday, November 15, at 4:00pm. Join in person at 2405 Siebel Center for Computer Science, 201 N. Goodwin Ave.

 

“Memory-Efficient Hashed Page Tables” and “SpecFaaS: Accelerating Serverless Applications with Speculative Function Execution”

Jovan Stojkovic

 

In this talk, he will discuss two of his recent works to be presented at HPCA 2023.

 

For the first work, we focus on hashed page tables (HPTs), which avoid the scalability challenges of traditional radix page tables by requiring only one memory access for address translation. Despite recent advances in HPT design, a major problem is their substantial requirement for contiguous memory, preventing them from widespread adoption. This work addresses HPTs’ contiguous memory needs by introducing the Logical to Physical (L2P) Table and the use of Dynamically-Changing Chunk Sizes. These techniques break down the HPT into memory-efficient, discontiguous physical-memory chunks. In addition, we also introduce two techniques that minimize HPTs’ total memory needs and, indirectly, the contiguity memory requirements. Compared to state-of-the-art HPTs, our solution reduces the average contiguous memory allocation needs by 92%, and improves the average performance by 8.9%.

 

For the second work, serverless computing environments promise to attain high resource utilization while applications run at low cost. However, they induce substantial execution overheads, especially in applications with many control- and data-dependent functions. We propose to accelerate serverless applications with a novel approach based on software-supported speculative execution of functions, called SpecFaas. In SpecFaaS, functions in an application are executed early, speculatively, before their control and data dependences are resolved. As a result, the execution of downstream functions is overlapped with that of upstream functions, substantially reducing the end-to-end execution time of applications. We prototype SpecFaaS on top of Apache OpenWhisk, and achieve an average speedup of 4.6X, increased application throughput by 3.9X, and a tail latency reduction of 58.7%