Understanding how microarchitecture impacts microservices

Title: Understanding how microarchitecture impacts microservices

Speakers: Ashraf Mahgoub, Harshad Sane, and Kshitij Doshi (Intel)

Abstract: Microservices offer numerous benefits, including scalability, modularity, and scheduling flexibility. Accordingly, organizations that adapt microservices architecture in their applications can handle higher traffic loads scalably and allocate resources effectively. However, the underlying microarchitecture plays a crucial role in microservices performance. The interaction between the microservices behavior, the underlying microarchitecture, and placement decisions impose various trade-offs. For instance, two microservices that have a high inter-communication traffic are better collocated on the same host to minimize communication latency. Obviously, this comes with reduced placement flexibility. On the other hand, if the two microservices contend for local resources (such as last level cache), spreading them on different hosts minimizes resource contention. In this tutorial, we describe a methodical approach for addressing this issue. Specifically, we recap `map-and-zoom` performance methodology in the light of microservices. The methodology pursues multiple objectives such as debugging performance regressions and identifying scaling bottlenecks. We show how a family of three tools (developed by Intel) can be used jointly to achieve agile bottlenecks detection as well as fast reaction times. Finally, we show how performance metrics counters can be collected (in a lightweight manner) for each process/container on a given host, analyzed in real-time, and depicted in time-line view.

Bios:

Ashraf Mahgoub is a cloud software engineer in the Cloud Engineering and Solutions Group. His technical expertise is in designing and implementing automated systems for resource management and performance optimization for cloud-native applications. He obtained his B.S and M.S degrees in Computer Engineering from Cairo University (2011 and 2015 respectively). He obtained his Ph.D. degrees in Computer Science from Purdue University (2022). His research interests span cloud-native systems, automated performance optimization, and self-aware & Adaptive Computing.

Harshad Sane is a performance engineer in the Data Center and AI group with a deep technical expertise in system software, memory, and CPU architectures. He specializes in performance monitoring, software optimization, and tool development with focus in the cloud domain. Harshad joined Intel as a College Graduate (RCG) in 2008 after completing his M.S. in Electrical and Computer Engineering from UC Boulder.

Kshitij Doshi works at Intel Corporation in the Data Center and AI group, where he focuses on performance optimization of workloads and cloud instances. He obtained his undergraduate degree in Electrical Engineering from IIT Mumbai (1982) and his M.S. and Ph.D. degrees in Electrical and Computer Engineering from Rice University (1985 and 1989, respectively). His research interests span distributed systems, memory and storage architectures, and resource management.