Re: Slow response from Squid beta 23...

On Friday 6 December 96, at 1 h 49, the keyboard of "Walter Klomp"
<walter@swiftech.com.sg> wrote:

> Why does it still take 2-5 seconds before I get actually a response from my
> proxy server?

There can be many reasons. Since it took me two months to tune my
Alpha/DU machine to get a reasonable response time, I'm hardly a good
consultant but let's try:

> Also the cache-size on the harddisk never grows over 1.1 Gb while
> I set it to 2.5 Gb.

In Squid 1.0, this was possible if your clients used the cache in a
moderate way. Either because they don't spend time with the Web or they
bypass the cache.

In 1.1, with a different expiration model, I'm not sure. I think that
Squid no longer deletes files is the cache is not full (can anyone
confirm?).

> My VM cache is full of expired gif's and jpg's (1970...) in
> my VM (after 10 min loading) there is not 1 document NOT expired... I use the

This is normal in Squid 1.1. Many documents are pre-expired (and will be
checked with a GET If-Modified-Since) and the age is the beginning of
times (Unix was born on 1st January 1970).

I agree with Jonathan Larmour on the tuning of refresh_pattern. Default
options render the cache less efficient: it always have to check the
validity of files with the remote Web server.

> default settings, what can I do to improve the response time?

Check the following:

- what malloc is used by SCO (or switch to a better operating system,
does anyone use SCO here?). Most malloc shipped with commercial Unix are
inefficient or broken (a colleague showed me a paper in Usenix 1986
describing a bug in malloc, bug which is still present in Digital Unix).
May be you can try GNU malloc.
- check paging. I know you have 192 Mb but Squid can eat a lot more quite
easily. Look at it with ps, use vmstat.

More info:

cachemgr's "info" should tell you interesting things. Post it here.

echoping can tell you if the problem is remote or local. Look at this
trace to a specific Web server:

% echoping -n 5 -h http://www.microsoft.com/ cache.cnrs.fr:3128
Elapsed time: 4.585736 seconds
Elapsed time: 23.056653 seconds
Elapsed time: 4.858977 seconds
Elapsed time: 2.103913 seconds
Elapsed time: 5.223965 seconds

See? Every request takes the same (long) time. This is not Squid's fault,
it is because Microsoft sends Expires fields which prevent caching. When
Squid just proxies, you actually measure the response time of the remote
Web server (which can be huge, even for small pages: TCP connections are
expensive over long lines).

On a correct Web servers such as www.nlanr.net, only the first time
should be long, if your Squid actually caches. If not, you have the
problem.
Received on Thu Dec 05 1996 - 12:03:39 MST