On 8/17/05, lokesh.khanna@accelonafrica.com
<lokesh.khanna@accelonafrica.com> wrote:
> What parameters do I need to check to get best performance and caching.
> My squid is currently handling 80 req/sec and during peak hours nearly
> 1500 users uses caching server.
That is a respectable traffic volume.
IMHO, the best thing you can do to improve a cache and the end user
experience is to add as much RAM as you can afford (as the box can take).
Another option might be to implement multiple Squid peers in parallel,
varying the object purge policy and max/min object sizes amongst the
different cache peers. I've only recently started to play with this idea,
now that I have six "parent" caches each with between 1-4GB of RAM
and 16-36GB of dedicated FCAL-attached cache disk storage.
> What parameter do I need to poll to check object hit rate?
> There are other parameters also like Byte Hit rate, request Hit ratio
> etc. How can I get detail on this. Is there any website for this?
You probably don't want to poll to get these statistics, but instead run
something like "calamaris" nightly or weekly to get the _average_ hit
rate over a longer time period.
> Is it really impossible to get better performance
> and more caching up to 30%
Your real world cache hit/byte ratios (are going to be for the most
part defined by two variables over which you have little or no control --
the requests sourced by your users, and how "cache friendly" are
the content (and headers) served up by the remote servers.
If you could force your customers to only ever visit three web sites
which provide purely static content with "Expires" and "Last-Modified"
headers, then your cache rate could be substantially higher :)
Realistically, aside from adding cache disk/mem to ensure the cache
doesn't have to toss out any data it might need later, the other tunables
available in squid can be risky at best. If you make adjustments to the
"freshness" calculations, your users may start to complain about getting
old stale copies of pages from the cache long after the source web site
has updated their content.
Kevin Kadow
Received on Wed Aug 17 2005 - 02:39:23 MDT
This archive was generated by hypermail pre-2.1.9 : Thu Sep 01 2005 - 12:00:02 MDT