Re: [squid-users] tuning an overloaded server

Chris Robertson wrote:
> Al - Image Hosting Services wrote:
>> Hi,
>>
>> I hope that someone can help me. I have 3 servers running squid acting
>> as a web accelerator for a single http server. I have been having a
>> large amount of problems with the http server locking up and having
>> drive failures and squid seems to be a great solution. I was
>> pleasantly surprised at the performance of squid when I first
>> installed it. Even on the slowest of my servers, it seemed faster than
>> directly from the http server, but after I put the load back on the
>> slowest of my servers it was 0% idle.
>
> Where is the CPU usage at, User, System or Wait?
>
>> Since this server had a second hard drive that was basically not being
>> used, I moved the squid cache over to it. I then did some googling and
>> I made some other changes and the server is now only 0% idle a few
>> times a day. I am really suprised at just how tunable squid is and it
>> was wonderful that a change in configuration could make such a
>> difference. But since the busy season is coming, I still think that
>> more can be done to tune it.
>>
>> Here are the changes that I made to the squid.conf:
>> redirect_rewrites_host_header off
>> acl all src 0.0.0.0/0.0.0.0
>> acl manager proto cache_object
>> acl localhost src 127.0.0.1/255.255.255.255
>> acl to_localhost dst 127.0.0.0/8
>> acl SSL_ports port 443
>> acl CONNECT method CONNECT
>> http_access allow manager localhost
>> http_access deny manager
>> http_access deny !Safe_ports
>> http_access deny CONNECT !SSL_ports
>> http_access allow all

Not a good idea, limit the access to just the domains you are serving
where at all possible.

>> icp_access allow all
>> http_port 80 accel vhost
>> cache_peer 10.10.1.4 parent 80 0 no-query originserver
>> hierarchy_stoplist cgi-bin ?
>> cache_mem 16 MB

cache_mem is worth expanding, it lets more objects stay in memory and
skip the disk IO waiting times.

>> maximum_object_size_in_memory 128 KB
>> memory_replacement_policy heap GDSF
>> cache_replacement_policy heap GDSF
>> cache_dir ufs /var/squid/cache 200 16 256
>> maximum_object_size 2048 KB
>> cache_swap_low 90
>> cache_swap_high 98
>> access_log /var/log/squid/access.log squid
>> cache_log /var/log/squid/cache.log
>> cache_store_log /var/log/squid/store.log

dump cache_store_log its a waste of disk IO unless you really need it
for something weird.

>> emulate_httpd_log on
>> buffered_logs on
>> acl QUERY urlpath_regex cgi-bin \?
>> cache deny QUERY
>> refresh_pattern ^ftp: 1440 20% 10080
>> refresh_pattern ^gopher: 1440 0% 1440
>> refresh_pattern . 0 20% 4320
>> acl apache rep_header Server ^Apache
>> broken_vary_encoding allow apache
>> half_closed_clients off
>> icp_port 0
>> log_icp_queries off
>> coredump_dir /var/squid/cache
>>
>> Are there any other changes that can be made to improve cpu usage?
>>
>> I also did an install from source, since I am using NetBSD systems I
>> used pkgsrc to install it. There are several build options:
>>
>> arp-acl
>> aufs
>> carp
>> coss
>> diskd
>> icmp
>> ipf-transparent
>> pam-helper
>> pf-transparent
>> snmp
>> ssl
>> unlinkd
>>
>> I looked up what some of these are, but I am not sure how they are
>> used by squid or if I really need them. My thought is that if I
>> compiled squid without these options it could improve performance. I
>> wish there was more information on running squid on BSD systems
>> (especially recent info) and then maybe I could be sure what effect
>> compiling without an option would have. For example, I know what pam
>> is, but do I need it?
>
> You are not using authentication, so could do without PAM.
>
>> I looked up what unlinkd does and I can see it running, but will it
>> help cpu usage to use it?
>
> Using --enable-truncate "gives a little performance improvement, but may
> cause problems when used with async I/O. Truncate uses more filesystem
> inodes than unlink.." asynch I/O refers to diskd, aufs and coss. The
> only way to see if that will lower your CPU usage is to try, but...
>
>> I don't see diskd running as a seperate process, so I don't know that
>> I am even using it.
>
> Your cache_dir line specifies "ufs". To use diskd, you need to change
> that line to...
>
> cache_dir diskd /var/squid/cache 200 16 256
>
> ... with a couple more optional parameters to tune outstanding
> requests. To use aufs use...
>
> cache_dir aufs /var/squid/cache 200 16 256
>
> Either one is likely to have a positive effect on your CPU utilization
> as Squid will spend less time waiting for I/O. Turning off the
> cache_store_log is probably not going to hurt anything. Making sure
> your origin server gives proper expiry information and removing the
> QUERY acl (and the "cache deny QUERY" line that relies on it) would
> remove one more CPU sink.

That said BSD family of systems get more out of diskd than aufs in
current Squid.

>
>> The only one that I am sure that I need is ssl.
>>
>> Any help would be greatly appreciated!
>>
>> Best Regards,
>> Al
>
> Chris

On top of all those adjustments if you still need a boots make sure you
are using the latest Squid-2.7. Those releases have the most performance
optimizations for high-throughput situations.

Amos

-- 
Please be using
   Current Stable Squid 2.7.STABLE5 or 3.0.STABLE10
   Current Beta Squid 3.1.0.2