Re: maximum_object_size_in_memory default

From: Amos Jeffries <squid3@dont-contact.us>
Date: Wed, 7 May 2008 16:20:19 +1200 (NZST)

> On Sat, 2008-05-03 at 00:02 +1200, Amos Jeffries wrote:
>> The recent squid-users request made me check the documentation for
>> maximum_object_size_in_memory. I realised that in RAM-only caches it
>> would limit the max-size of all files.
>>
>> Given that RAM-cache is now the backup default method. And the default
>> size for maximum_object_size_in_memory is currently 8KB, it would be a
>> drag for RAM-caching setups.
>>
>> Votes on raising that to at least 64KB to make RAM-caches reasonable
>> HIT-able?
>>
>> 128KB?
>
> With 8MB default value for cache_mem, memory cache is probably a poor
> backup regardless for the maximum_object_size_in_memory and smaller
> object size values are better when the cache size is so small.
>
> Perhaps maximum_object_size_in_memory default should be computed based
> on cache_mem? We could try to set maximum_object_size_in_memory to cache
> 10,000 objects, with 8KB minimum and 64KB maximum, for example.

If we were going to set an arbitrary number of ideal objects, I think
setting a max-value would be a mistake. We could add an auto[=N] option if
we are going to get fancy and do calculations. It is after all a
theoretical max, objects will match their own distribution in live
traffic.

Min is a good idea though. At current settings default Mem-cache would
hold at least 5k objects. Just one much smaller than the average image and
page size.

However this is just a policy-change vote. Not a full-feature addition at
this point (leave that till 3.2+). cache_mem policy may need changing also
despite me not requesting it initially.

>
> HTH,
>
> Alex.
> P.S. It would be nice to update cache_mem docs to point to
> maximum_object_size_in_memory.

Easier done than said ;-)
I agree it should be mentioned in whatever replaces the obsolete
null-store docs.

Amos
Received on Wed May 07 2008 - 04:20:24 MDT

This archive was generated by hypermail 2.2.0 : Tue May 13 2008 - 12:00:04 MDT