Split gRPC: An Isolation Architecture for RPC Software Stacks

Remote procedure calls are a major contributor to performance variance in distributed systems due to lack of isolation and contention on shared resources. We propose a novel split architecture for a popular RPC framework, gRPC. The design partitions RPC applications into two communicating components: one dedicated to user-implemented business logic, and one dedicated to RPC infrastructure processing. The infrastructure process can be run on a dedicated core or on a smart NIC (e.g., IPU or DPU), providing effective physical isolation and predictable performance. An initial evaluation shows that the split architecture adds modest overhead for average case latency, but allows for lower latency and and higher throughput under host CPU load.

Authors

Institutions

Publication Details

Published
2024-08-29
DOI
https://doi.org/10.1145/3678015.3680484
Primary Topic
Advanced Data Storage Technologies
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Split gRPC: An Isolation Architecture for RPC Software Stacks

Rajit Manohar, Pietro Bressana, Rui Li, Robert Soulé et al.
Advanced Data Storage Technologies
article

Split gRPC: An Isolation Architecture for RPC Software Stacks

Rajit Manohar, Pietro Bressana, Rui Li, Robert Soulé, Edmund Chen, Peter Alvaro, James Tsai, Uri Cummings, Esteban Ramos
article en

Abstract

Remote procedure calls are a major contributor to performance variance in distributed systems due to lack of isolation and contention on shared resources. We propose a novel split architecture for a popular RPC framework, gRPC. The design partitions RPC applications into two communicating components: one dedicated to user-implemented business logic, and one dedicated to RPC infrastructure processing. The infrastructure process can be run on a dedicated core or on a smart NIC (e.g., IPU or DPU), providing effective physical isolation and predictable performance. An initial evaluation shows that the split architecture adds modest overhead for average case latency, but allows for lower latency and and higher throughput under host CPU load.

Intel (United States) (US), University of California, Santa Cruz (US), Yale University (US), Santa Cruz County Office of Education (US)
Openalex Percentile: Top 85%
Advanced Data Storage Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.