Intel Optane DCPMM and Serverless Computing

TL;DR

This paper explores the performance of Intel Optane DCPMM in memory mode for scalable graph processing and examines execution delays in serverless computing, highlighting potential improvements with persistent memory.

Contribution

It provides an analysis of Optane DCPMM's scalability for memory-intensive applications and investigates serverless function delays, proposing future enhancements with persistent memory.

Findings

Optane DCPMM enables scalable graph processing on a single node.

Serverless function execution delays are affected by cold starts and concurrency.

Persistent memory could improve serverless performance in cloud platforms.

Abstract

This report describes 1) how we use Intel's Optane DCPMM in the memory Mode. We investigate the the scalability of applications on a single Optane machine, using Subgraph counting as memory-intensive graph problem. We test with various input graph and subtemplate sizes to determine its performance for different memory and CPU loads, as well as a comparison of performance on a single node Optane with a distributed set of nodes in a cluster using MPI. 2) We investigate the end-to-end execution delays in serverless computing and study concurrent function executions with cold starts. In future work, we will show that persistent memory machines may significantly improve concurrent function invocations in serverless computing including Amazon Lambda, Microsoft Azure Functions, Google Cloud Functions and IBM Cloud Functions (Apache OpenWhisk).