Re: [squid-users] issue with one of joe coopers modifcations

Hmmm... 26GB is close to the limit for a 512MB machine. But not over
it. You should be fine wrt memory.

What bandwidth are you supporting with this box? Are you overloading
the box or perhaps your DNS, such that requests in the queue grow too large.

2.2.14 has no memory bugs that I'm aware of (we have had units in the
field in the past running this version with no problems). But 2.2.14
does have security problems, and should probably be upgraded (Red Hat
has RPMs on their site--just follow their instructions for updating a
kernel).

What does your Squid process size look like when under load? I would
expect it to be about 300MB. Where is the rest of it going on your system?

Greg wrote:

> just so you the machine
> its 512mb ram, linux 2.2.14-50
> pentium II 333
> ultra 160 30 gig scsi drive
> handles about 500 to 1000 modem users and 6 lan users
>
> cache size is 26 gig.
> 30x10=300
>
> so that means that i have 212 meg left or close to that figure, now i know
> linux will use up a bit of memory
> but surely i wouldn't need more than 512 meg of ram???
> Thanks
> Greg
>
> i noticed you use 2.2.16, is 2.2.16 more stable and better than 2.2.14-5.0??
> Thanks
>
>
> ----- Original Message -----
> From: "Joe Cooper" <joe@swelltech.com>
> To: "Greg" <squid@australis.com.au>
> Cc: "squid" <squid-users@squid-cache.org>
> Sent: Friday, March 23, 2001 6:23 PM
> Subject: Re: [squid-users] issue with one of joe coopers modifcations
>
>
>
>> Hi Greg,
>>
>> Using my instructions has nothing to do with having too little memory in
>> your machine to handle a cache that size. 73 L1 directories??? Are you
>> really using a cache_dir that large?
>>
>> You are simply filling up your RAM with an in-core index of the cache
>> contents. This is normal behavior--Squid keeps a record of every object
>> in the store in memory. If your store is gigantic (as yours clearly
>> is), and your memory is not gigantic to match, you will run out of
>> memory. There is no leak, and there is no flaw in the malloc used by
>> default in Red Hat 6.2.
>>
>> Lower your cache_dirs to something sensible (1GB for every 10MB of RAM
>> is a safe number for a standard Squid compile--a little more RAM is
>> needed for an async i/o compile). This too, is covered in my
>> documentation for tuning Squid on Linux.
>>
>> Hope this helps.
>>
>> Greg wrote:
>>
>>
>>> Hello.
>>>
>>> I changed the default first level cache directory from default 14 and
>>> used his formula, i got 73, anyway getting to the point, basically what
>>> happens after about 3 to 4 weeks it uses up all the memory and hits swap
>>> space, i have tried, rebooting, no difference, and using the kill
>>> command, (had httpd and cron running, thought they were bad, so i killed
>>
>
>>> them) its still made no difference, so i am now using the alernative
>>> malloc (configure --enable-dlmalloc)
>>>
>>> and seeing if there is any other difference, the only thing i can't do
>>> is make a custom kernel (i think there are compiler problems in my
>>> version of redhat 6.2)
>>>
>>> thanks
>>>
>>> Greg
>>
>> --
>> Joe Cooper <joe@swelltech.com>
>> Affordable Web Caching Proxy Appliances
>> http://www.swelltech.com
>>

                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com
Received on Fri Mar 23 2001 - 16:11:27 MST