3.1.2 Matrix Server Administration Guide

Chapter 16: Performance Monitoring 208

kernel attempts to always use memory, but in ways that improve system

performance and can easily be reclaimed, such as a disk buffer cache. For

this counter, lower numbers are better because they indicate that the

kernel is under less pressure to balance memory resources.

Average swap memory utilization (%). This counter reports the

percentage of allocated swap space that is currently in use by the kernel’s

virtual memory subsystem. A high number for this counter could

indicate that you need more memory. It might also mean that you need a

bigger swap file, as the counter is just reporting a percentage. If the value

for this counter approaches 100%, the kernel will not be able to allocate

additional memory for applications, which can cause operations to fail.

Total filesystem I/O transfer rate (MB/s). This counter includes both

READ and WRITE I/O operations and specifies how much data was

actually transferred. PSFS filesystem I/O includes both meta-data

(internal filesystem I/O traffic) and application traffic. Some mount

options and workloads can result in much more meta-data and

application I/O than is observable from the application. If performance is

not meeting expectations, this is almost always a good place to look.

Total filesystem I/O operations (IO OPs/sec). This counter reports the

number of individual read or write requests that were serviced. This

value is related to the total filesystem I/O transfer rate, although typically

in an application-specific manner.

Average one minute run-queue depth. This counter specifies the depth

of the kernel’s process run-queue. It is a measure of how many processes

on average were waiting to run for the last minute at a given point in

time. This measure is also known as “load average,” but is more precisely

defined as ”run queue depth.” A lower value for this counter is better

because it means that the kernel and CPUs are able to keep up with the

demands for CPU time being made by the processes running on the

system.