Acceleration of latency-sensitive applications in virtualized environments

• Grant number: 23/00812-7
• Support Opportunities: Regular Research Grants
• Duration: September 01, 2023 - August 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Convênio/Acordo: MCTI/MC
• Principal Investigator: Ronaldo Alves Ferreira
• Grantee: Ronaldo Alves Ferreira
• Host Institution: Faculdade de Computação. Universidade Federal de Mato Grosso do Sul (UFMS). Campo Grande , SP, Brazil
• Associated researchers: Italo Fernando Scotá Cunha

Abstract

Cloud applications are complex and usually decomposed into microservices that must run at very low latencies not to impact their performance. However, microservices generally run in containers and incur high latencies when using the protocol stack of a general-purpose operating system (OS) like Linux. Recent works deal with this problem in two ways. Some seek to optimize transport protocols to take advantage of services and tools offered by the OS, while others develop new protocol stacks that run in user space and use kernel bypass techniques to avoid OS overheads. However, both approaches still have shortcomings that lead to poor performance and delays in request processing. In this project, we will address two fundamental problems to support applications that are latency sensitive in virtualized environments: (i) how transport protocol parameters interfere in the performance of applications in virtualized environments; and (ii) how to decrease application latency on multicore systems using kernel bypass. For the first problem, we will investigate methods to understand the effects of parameters of legacy transport protocols, such as TCP and QUIC, considering the requirements of different application types and network scenarios. For the second problem, we will explore and evaluate different protocol stack architectures that use kernel bypass on multicore servers, develop scheduling algorithms for systems that must operate in microseconds scale, and investigate how programmable network adapters can help accelerate latency-sensitive applications in virtualized environments. (AU)

Publicações científicas

(Referências obtidas automaticamente do Web of Science e do SciELO, por meio da informação sobre o financiamento pela FAPESP e o número do processo correspondente, incluída na publicação pelos autores)

CARVALHO, FABRICIO B.; FERREIRA, RONALDO A.; CUNHA, ITALO; VIEIRA, MARCOS A. M.; RAMANATHAN, MURALI K.. State Disaggregation for Dynamic Scaling of Network Functions. IEEE-ACM TRANSACTIONS ON NETWORKING, v. N/A, p. 15-pg., 2023-06-12. (15/24485-9, 20/05183-0, 23/00812-7, 14/50937-1)