Re: [squid-users] Squid Performance Issues - reproduced

From: Mika Aleksandroff <mika.aleksandroff@dont-contact.us>
Date: Wed, 1 Jan 2003 20:08:35 +0200

On Wed January 1 2003 17:19, Henrik Nordstrom wrote:
> Stephan Austermuehle wrote:
> > Configuration:
> > '--enable-removal-policies=heap lru' --enable-storeio=ufs,aufs,diskd
> > --disable-poll
> You should add --enable-pthreads in the mix, or apply the required
> patches. Many bad things will happen if you use --enable-async-io or
> --enable-storeio=aufs without --enable-pthreads in 2.5.STABLE1.

Strangest thing, I've been experimenting with this thing for a while now
and I am actually experiencing the exact same thing. I even downloaded and
compiled the daily (20021231) package and enabled pthreads to no apparent
effect. I've used 'wget -O /dev/null URL' for testing.

Now, what I have:
- Linux 2.4.20 SMP custom kernel
- Dual Athlon XP 1600+
- Adaptec AHA-3960D U160 /w 3 x IBM DDYS-T18350N
- 3 cache_dirs on separate disks, reiserfs
- Two identical Squid processes, one using aufs, the other ufs. Each drive
has 6G spool directory for each process (yes, I know, I don't have enough
drives...)

What I did:
- Created a 10 MB file on our local server and downloaded it several times
on different computers in same and different subnets on a 100M switched
network.

Results:
- The first time I get some 2+ MB/s transfer rate from the web server. An
example:
1041438822.418 5225 <client> TCP_MISS/200 10486044 GET <URL> -
DIRECT/<server> text/plain
- After that transfer rate (aufs) is always between 500-700 kB/s. From ufs
the transfer rate varies from 7.5 to 11 MB/s depending on client location
(logical/physical) in the network.
- When I increase the maximum memory object size to 16M (for TCP_MEM_HIT) I
get transfer rates of 7-11 MB/s from both Squids.

Now this is strange. But to make it even stranger I tried the old 2.4S6
(with aufs) as well and the result was even worse (200+ kB/s). Although
this may also be somewhat due to ridiculously low cache_mem (6M) which I
used. I guess the slight increase in object size when retrieved from cache
is normal?

1041441729.539 1351 <client> TCP_MISS/200 10486044 GET <URL> -
DIRECT/<server> text/plain
1041441776.570 42623 <client> TCP_HIT/200 10486051 GET <URL> - NONE/-
text/plain

Anyway, surely someone would have noticed it before if it was an actual bug
in Squid? I could do some further testing, if necessary, on 2.2 vs. 2.4
kernels and such but that would be on a slightly different hardware.

-- 
Mika Aleksandroff, Kymenlaakson AMK
________________________________________________
Words are a deceit, a shore no beach.
Received on Wed Jan 01 2003 - 11:08:38 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:12:24 MST