Jaeho Yang wrote:
> When the requests are 40~50 per seconds, the meter of cpu load
> increases upto 10.0 ~ 16.0. The squid uses the most cpu time,
> and the most cpu time was system mode (not user mode).
>
> The odd thing is that there are still 15~30% cpu idle when the
> cpu load indicates above 10.0.
load is NOT cpu load. %idle CPU is CPU load.
What you describe is the expected behaviour of a Squid compiled with
async-io. The reason why the load indication rises high is that more
threads gets activated to drive the disks. Translated to Squid behaviour
a load of ~10 is approximately 10 blocking disk operations on average.
If you want to compare Squid 1.1.X load and Squid 1.2beta async-io load,
the only number that you can use is %idle while processing the same
number of requests/second cached in the same way and nothing else going
on in the machine at the same time. I prefer to measure
requestrate/second when overloaded instead of CPU usage.
If your machine is low on CPU then async-io is not of any use (it may
actually make things worse in a low CPU situation). async-io is useful
when disk I/O is a bottleneck, which quickly is the case when you have
more than one disk and a fairly powerful CPU (pentium 133 and up).
--- Henrik Nordström Sparetime Squid HackerReceived on Sat Jun 20 1998 - 09:34:24 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:40:46 MST