RE: Increasing MEM_HIT in Squid: just add RAM?

From: Kendall Lister <kendall@dont-contact.us>
Date: Fri, 14 Jan 2000 12:04:29 +1100 (EST)

On Thu, 13 Jan 2000, Steve Snyder wrote:

> 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.

The intuition you're using here assumes that multiple requests happen soon
after the first request; I certainly don't want to stop you from
increasing your 'cache_mem', but in the long run it is unlikely to gain as
much as one might imagine. However, the ultimate test is of course whether
it improves your web surfing experience, and since it does, you should do
it. Heed the N/3 rule of thumb, and go for it.

> 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?

This is what was referred to in the first response to your question - due
to blocking versus non-blocking system calls, it can be more efficient to
let the operating system handle page faults than asking Squid to do it.

--
 Kendall Lister, Systems Operator for Charon I.S. - kendall@charon.net.au
  Charon Information Services - Friendly, Cheap Melbourne ISP: 9589 7781
Received on Thu Jan 13 2000 - 18:13:12 MST

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