Re: [squid-users] Performance tuning Squid box for ISP traffic

Hello Ow,

>>I haven't tested calamaris, but it seems unsuitable for me, because my
>>access logs (with log_mime_hdrs on) grow to 3 GB during 5 hours, so I
>
> Wow. That certainly is a lot, but since you have log_mime_headers.
> That sort of explains it.
>
> Can I ask why do you need it? isn't it for problem tracking only?

Exactly. For expample, one client complained about Squid because she
used HTTP/0.9! So I need to be ready to see what has happend when there
is an error report.

>>>2) Any fine tune parameters for better performance rather than using default
>>>values in squid.conf ?
>>
>>- diskd instead of ufs
>
> This is FreeBSD specific right?
Yes. I'm running Linux so I have aufs, but the guy asked about FreeBSD
for which, I believe, the diskd should be better.

>>- httpd_accel_with_proxy off (you are intercepting, right?)
>
> Do you see problems running in intercept mode? I've read there are quite
> a number of drawbacks with this.

You mean disadwantages with running in intercept mode with
"httpd_accel_with_proxy off", i.e. without a possibility to use the
proxy mode? I cannot see any.

If you are in intercept mode, you don't need the proxy mode. I am just
carefull, I don't want others exploiting my Squid which is far away from
perfect configuration :-)

>>And one more remark. People in this list keep saing the reiserfs is the
>>best. I decided to use ext3, anyway.
>
> I use reiserfs on a 30GB cache.
Can you somehow measure your Thoughput? Sometimes I am watching the load
going over 3, I believe the bottleneck are the disks, but I am not realy
sure.
Today I changed 3 X 28 GB cache from ext2fs to ext3fs and I haven't
noticed any performance change so far (good :-)

>>According to the performance benchmark in the Duane Wessels book "Squid
>>the definitive guide", ufs with reiserfs(notail, noatime) has only 61%
>>Throughput of ufs with ext3fs(event without notaime option).
>>Does anyone have a comment to this?
>
>
> I have (just bought <1 1/2 week ago) and I've yet to reach that page.
> But yeah, you're right, based on his test methods:
>
> on Linux 2.4 with 32 threads, ext3fs is fastest.
>
> scheme FS Mount Opt throughput
> aufs ext3 noatime 168
> ufs reiserfs noatime,notail 21.4

Ooops, you are comparing aufs with ufs scheme, that is not fair to
reiserfs :-)

Have a nice day, Ow,
Marji
Received on Thu Dec 09 2004 - 01:47:20 MST