Re: [squid-users] Recommended RAM

If I were configuring such a box I would use 24GB per disk for cache_dir
(more is overkill for your bandwidth load and will slow performance
somewhat). Even 24GB is an awful lot for the client load you're
supporting. You're retention will be quite a bit longer than is really
needed. But there's no real harm in that, as long as you don't run out
of memory.

Your cache is probably on the borderline of being busy enough to justify
async i/o, and the fact that the disks are IDE may cause you to find
that throughput is limited by the disk blocking of a single process
Squid. What speed is your processor? Async i/o is much more demanding
on the processor than a single process Squid. I'll assume since you've
got a ton of RAM, you haven't skimped on the processor--if you've got
600MHz or more in the CPU you should be just fine with an async compile.

All that being said, 2000/min is only 33/sec, which any kind of Squid
compile can handle if given two 7200 RPM disks (or even 5400 RPM disks
with reasonable seek times).

Good luck!

Shane T. Ferguson wrote:

> sorry, yes that's what I meant 2000 req/minute.
>
> How should I lay out my cache_dir (using 1GB of RAM limit) to maximize my
> performance with the 2x 40Gb drives?
>
> Also, how much load is considered a 'busy cache' to enable async-io?
>
> Thanks for the advice!
>
> Shane
>
>
>
> -----Original Message-----
> From: Joe Cooper [mailto:joe@swelltech.com]
> Sent: Monday, April 23, 2001 9:31 PM
> To: Shane T. Ferguson
> Cc: squid-users@squid-cache.org
> Subject: Re: [squid-users] Recommended RAM
>
>
> 2000 reqs/sec? So you're supporting a 155Mbps link?
>
> A single Squid box will not do 2000reqs/sec no matter what you do. Very
> few web caches will, and certainly not from 2 IDE disks (at the
> cacheoff, you'll see boxes with 16 SCSI Ultra 160 disks doing about that
> rate--and they aren't running Squid).
>
> I'll guess you mean 2000/min?
>
> If so, that's not too hard at all even without an async i/o compile.
> But you don't have enough RAM for 70GB of cache_dir space. 1GB of RAM
> /might/ make it. I would actually recommend you drop the size of your
> cache_dirs some, and raise the RAM some as well.
>
> Shane T. Ferguson wrote:
>
>
>> Hi,
>>
>> What would be the recommended amount of RAM for Squid 2.4.STABLE1 on Linux
>> 2.2.18
>> running transparently (WCCP) with a Cisco router. For disk drives, I am
>> using 2x 40GB ATA-100 drives set up with the following in squid.conf:
>>
>> cache_dir ufs /cache2 35000 32 256
>> cache_dir ufs /cache3 35000 32 256
>> cache_mem 32 MB
>> memory_pools off
>> maximum_object_size 10240 KB
>> maximum_object_size_in_memory 1024 KB
>>
>> The average client load is approximately 2000 requests/second and will be
>> increasing over the next month or so.
>>
>> The reason I ask about RAM requirements is because I currently have 768MB
>> installed and an enormous amount of paging is occuring (running out of
>
> swap
>
>> space
>> fast). The problem fits the description of the FAQ section:
>> 11.17 My Squid becomes very slow after it has been running for some time.
>>
>> I have yet to try GNU malloc. If anyone can let me know if I'm actually
>> running
>> this box with insufficient RAM, i'd appreciate it. One other questions is
>> should a box with this load have been set up with asyn-io enabled?
>> Thanks
>>
>> Shane T. Ferguson

                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com
Received on Mon Apr 23 2001 - 21:58:54 MDT