Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Squeezy: Rapid VM Memory Reclamation for Serverless Functions

Orestis Lagkas Nikolos, Chloe Alverti, Stratos Psomadakis, Georgios Goumas, Nectarios Koziris

TL;DR

The paper tackles memory elasticity for N:1 FaaS VMs by exposing and exploiting the predictability of function memory footprints. It introduces Squeezy, an OS-level extension that partitions guest memory into fixed chunks per function and a shared partition, enabling fast, migration-free memory reclamation during function termination. Implemented in Linux 6.6 and integrated with an OpenWhisk-based runtime, Squeezy delivers sub-second reclamation of multiple GiB, maintains low CPU overhead, and keeps tail latency bounded under realistic serverless workloads. Compared with baseline approaches and 1:1 microVMs, Squeezy reduces memory waste and improves end-to-end performance, making memory elasticity viable at sub-second timescales for serverless deployments.

Abstract

Resource elasticity is one of the key defining characteristics of the Function-as-a-Service (FaaS) serverless computing paradigm. While compute resources assigned to VM-sandboxed functions can be seamlessly adjusted on the fly, memory elasticity remains challenging. Hot(un)plugging memory resources suffers from long reclamation latencies and occupies valuable CPU resources. We identify the obliviousness of the OS memory manager to the hotplugged memory as the key issue hindering hot-unplug performance, and design Squeezy, a novel approach for fast and efficient VM memory hot(un)plug, targeting VM-sandboxed serverless functions. Our key insight is that by segregating hotplugged memory regions from regular VM memory, we are able to bound the lifetime of allocations within these regions thus enabling their fast and efficient reclamation. We implement Squeezy in Linux v6.6 as an extension to the OS memory manager. Our evaluation reveals that Squeezy is an order-of-magnitude faster than state-of-the-art, keeping tail latency bounded, when reclaiming VM memory, achieving sub-second reclamation of multiple GiBs of memory while serving realistic FaaS load.

Squeezy: Rapid VM Memory Reclamation for Serverless Functions

TL;DR

The paper tackles memory elasticity for N:1 FaaS VMs by exposing and exploiting the predictability of function memory footprints. It introduces Squeezy, an OS-level extension that partitions guest memory into fixed chunks per function and a shared partition, enabling fast, migration-free memory reclamation during function termination. Implemented in Linux 6.6 and integrated with an OpenWhisk-based runtime, Squeezy delivers sub-second reclamation of multiple GiB, maintains low CPU overhead, and keeps tail latency bounded under realistic serverless workloads. Compared with baseline approaches and 1:1 microVMs, Squeezy reduces memory waste and improves end-to-end performance, making memory elasticity viable at sub-second timescales for serverless deployments.

Abstract

Resource elasticity is one of the key defining characteristics of the Function-as-a-Service (FaaS) serverless computing paradigm. While compute resources assigned to VM-sandboxed functions can be seamlessly adjusted on the fly, memory elasticity remains challenging. Hot(un)plugging memory resources suffers from long reclamation latencies and occupies valuable CPU resources. We identify the obliviousness of the OS memory manager to the hotplugged memory as the key issue hindering hot-unplug performance, and design Squeezy, a novel approach for fast and efficient VM memory hot(un)plug, targeting VM-sandboxed serverless functions. Our key insight is that by segregating hotplugged memory regions from regular VM memory, we are able to bound the lifetime of allocations within these regions thus enabling their fast and efficient reclamation. We implement Squeezy in Linux v6.6 as an extension to the OS memory manager. Our evaluation reveals that Squeezy is an order-of-magnitude faster than state-of-the-art, keeping tail latency bounded, when reclaiming VM memory, achieving sub-second reclamation of multiple GiBs of memory while serving realistic FaaS load.

Paper Structure

This paper contains 22 sections, 11 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Function Isolation Models
  4. Dynamically Adjusting VM Memory
  5. Squeezy Overview
  6. Squeezy Implementation
  7. Squeezy OS Mechanisms
  8. Squeezy Integration Into a FaaS Runtime
  9. Methodology
  10. Experimental Setup
  11. Evaluation Scenarios
  12. Evaluation
  13. Reclaiming VM Memory
  14. Reclamation Latency
  15. CPU Utilization.
  16. ...and 7 more sections

Figures (11)

  • Figure 1: The N:1 model reserves memory for N instances even when the load is low and a large number of concurrent instances are reclaimed by the FaaS runtime.
  • Figure 2: An analysis of the invocation pattern of the 10 most popular functions in the Azure production traces azuretrace shows that thousands of instances are created and evicted per minute. Mechanisms that enable the dynamic and agile resizing of N:1 VMs are essential to save resources.
  • Figure 3: Linux interleaves processes footprint within physical memory blocks. When F2 exits, the F1 pages that are co-located in the same blocks have to be migrated to reclaim memory. Squeezy partitions processes footprints.
  • Figure 4: Squeezy integration into a serverless runtime
  • Figure 5: Average latency (ms) to reclaim memory of different sizes from a guest with memhog-loaded CPUs.
  • ...and 6 more figures