Thrashing Guide

Introduction

Thrashing is an undesirable state where most of the processes in the system are in the process of paging-in memory and none of the threads are doing useful work. This results in unacceptable system performance.

Background information

These topics are useful background information on thrashing:

• Paging

• Virtual memory

Thrashing features

The signs of thrashing in demand paging are:

• The system performance rapidly becomes unacceptable, since no threads are doing useful work. The device will become unresponsive or appear to 'hang'.

• When observing paging activity logs it's seen that the same set of pages are paged in and out many times over the course of a use case.

• When observing paging activity logs there are long periods where many threads are waiting for pages to be paged in.

• There are large periods of null thread activity.

Thrashing prevention

The following is a means of preventing thrashing from occurring:

• Increase the size of the paging cache. This reduces page faults and hence the need to page-in memory.

• Mark the code or data involved as unpaged, for example if a 20MB buffer is actively used through a use-case and discarded afterwards there is little use in making it paged as it will need to always be in the page cache.

• Reduce the working set size so that it fits into the paging cache, for example instead of having four activities running concurrently, serialize them.