Re: Is squid unable to handle the load?

From: Markus Storm <Markus.Storm@dont-contact.us>
Date: Wed, 27 May 1998 09:36:11 +0200

Mark Dabrowski wrote:
>
> Ok, here's what I've done:
>
> We have ordered 3 new Adaptec 2940 U2W controllers (UltraWide 2) which go
> 80MB/s, but unfortunately couldn't find anywhere 4GB Ultra Wide 2 drives
> and also BSD 3.1 does not recognize (support) 2940U2W at all ...
>
> So we ended up installing 2 more 2940 UW's and adding 2 x 4GB drives.

You really don't need fast (Ultra, Wide) or lots of controllers. A single
Fast/Wide one does for us (currently maxing out the CPU at around 30-40
requests/second).
Load distribution is the key. Static (using multiple cache_dirs) is good,
dynamic (which you get when you stripe disks together) is even better.

>
> The disk configuration of Squid box is 3 AHA 2940 UW controllers and 1 X
> 9GB Drive and 2 X 4.5GB drives. I have defined 3 equal partitions in
> cache_dir, each on each disk.
>
> Since we just finished installation, i didn't have a chance to make any
> tests, but when the system is almost idle and only 5-10 clients use squid
> there is something weird happening... Every 15-20 seconds I would notice
> number of disk operations/sec jump to about 100 on each disk (which causes
> cpu idle to go down to 100%) and then resumes normally.
>
> Is squid doing some disk operations every 15-20 seconds? If yes, under
> normal conditions I would not think it can halt the system.... but it
> does... Any ideas?

Assuming you haven't changed your config (in particular cache_mem_low/high),
then most likely UNIX does.
Depending on your UNIX variant (not sure about BSD), part or all of the
filesystem
or VM cache is flushed to disk.
Try something like running a script doin' while (true) { sync; sleep 2 }.

Markus
Received on Wed May 27 1998 - 00:44:48 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:40:20 MST