RE: Increasing MEM_HIT in Squid: just add RAM?

From: Steve Snyder <swsnyder@dont-contact.us>
Date: Thu, 13 Jan 2000 14:00:45 -0500 (EST)

On Thu, 13 Jan 2000 16:45:23 -0000, Dave J Woolley wrote:

>> From: Steve Snyder [SMTP:swsnyder@home.com]
>>
>> I want to bump up the percentage of MEM_HIT hits in Squid
>> v2.2STABLE5.
>>
> Why? What OS? How is Squid used on the
> machine.
>
> Particularly for Linux this can reduce
> performance by causing the process to
> block on page faults whereas blocks not
> in the OS disk cache do not force
> single threading on Squid. The OS disk
> cache is still memory.

Yes, this for a Linux system. Why? Because MEM_HIT is so much faster
than TCP_HIT. Besides being intuitively obvious, analysis of the
Squid logs confirms this.

Of course there is no advantage to increasing "mem_cache" if it
results in an increase in swapping to disk. My question presupposes
that the machine has the RAM to accommodate an increase in the size of
Squid.

Regarding the disk cache being memory also, true enough. But surely
it is faster to have Squid satisfy a request from its own memory than
to have it call the operating system?

*** Steve Snyder ***
Received on Thu Jan 13 2000 - 12:10:02 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:50:22 MST