Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
Pournaras Georgios
Supervisors info:
Βασίλειος Καρακώστας, Επίκουρος Καθηγητής, Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Original Title:
Benchmarking Support for RISC-V CPUs in Serverless Computing
Translated title:
Benchmarking Support for RISC-V CPUs in Serverless Computing
Summary:
Serverless computing has emerged as a competitive cloud computing paradigm. At the same time, the open-source RISC-V ISA has gained a lot interest and the first RISC-V systems have already started to appear in the server market for datacenter and cloud computing. The combination of these two computing trends necessitates the performance assessment of the impact of the RISC-V ISA and relevant processor implementations when executing serverless workloads, particularly with respect to well-established ISAs and processor designs. However, currently, there is no benchmarking support for systematically evaluating serverless workloads on RISC-V systems. The goal of this thesis is to bridge this gap in benchmarking support across the layers of the computing stack, from the microarchitecture up to the application. We rely on vHive’s vSwarm that is a recently proposed serverless benchmark suite, and on vSwarm-u that provides infrastructure for executing serverless workloads in the gem5 microarchitecural simulator. We port several workloads from vSwarm to the RISC-V ISA enabling their execution on RISC-V systems. We also enhance vSwarm-u to enable the execution of those serverless workloads on simulated RISC-V CPUs using gem5. To achieve our goal, we address several challenges that mostly stem from the immaturity of the RISC-V software ecosystem. To demonstrate the usefulness of the enhanced benchmarking infrastructure, we evaluate the execution of the ported serverless workloads on a simulated RISC-V out-of-order multicore system. We also compare its execution with an equivalent x86 system. Our evaluation results highlight the important performance trade-off of cold vs warm execution for serverless workloads. Overall, our contributions pave the way for further experimentation with serverless workloads on RISC-V platforms, as well as for further comparison across various ISAs and processor microarchitectural parameters.
Main subject category:
Technology - Computer science
Keywords:
Serverless computing, benchmarking, RISC-V, software porting, vSwarm, docker, gem5, microservices