QStack: Re-architecting User-space Network Stack to Optimize CPU Efficiency and Service Quality

TL;DR

QStack is a flexible user-space network stack architecture that improves CPU efficiency and reduces tail latency in datacenter front-end servers by enabling elastic resource management and fine-grained prioritization.

Contribution

The paper introduces QStack, a novel user-space network stack design that balances CPU efficiency and low latency through elastic framework and application-defined priorities.

Findings

QStack outperforms existing user-space network stacks in CPU utilization.

QStack achieves lower tail latency in datacenter server tests.

QStack demonstrates effective resource collaboration across CPU cores and network layers.

Abstract

TCP/IP network stack is irreplaceable for Web services in datacenter front-end servers, and the demand for which is growing rapidly for emerging high concurrency network service applications (including Internet, Internet of Things, mobile Internet, etc.) especially. The existing network stack schemes often face the dilemma between the data center server resource utilization (i.e., high CPU efficiency) and application service quality (i.e., low tail latency). We break this dilemma via a flexible architectural design QStack, which simultaneously achieves CPU efficiency and low tail latency in user-space network stack for front-end datacenter server. QStack proposes elastic framework and application definable full-datapath priority, such that the network stack collaboration among CPU cores horizontally and coordination across network layers in fine grained vertically on demand. We prototype QStack on commodity servers. Testbed experiments demonstrate the effectiveness of QStack over state-of-the-art user-space network stack designs.