Re: [squid-users] CPU utilization performance issue

Schelstraete Bart Fri, 08 Aug 2003 12:26:48 -0700

I know that ext3 uses a lot more CPU on a lot of disk I/O. This is a known issue. Btw, did you compile the Squid with the gnu-regex option? Because that one is faster on Linux systems. (if you're using a lot of acl's)


rgrds,
         BArt
Tay Teck Wee wrote:

Hi Bart,

I'm using reiserfs. aufs coz its more suitable for linux.
--
Wolf
--- Schelstraete Bart <[EMAIL PROTECTED]> wrote:
Hzllo,

Why not using Reiser instead of ext3 with diskd? I read a lot of articles that are saying that reiser is much fast for Squid. (a lot of 'small files')

Bart

Zand, Nooshin wrote:

Hi, I am just wonder why you are not using diskd. Based on Benchmarking that I read, diskd provides

faster I/O performance.

I am planning to run squid on Linux Redhat 9.0 and

thinking to use ext3 and diskd.
Thanks,
Nooshin
-----Original Message----- From: Tay Teck Wee
[mailto:[EMAIL PROTECTED]

Sent: Friday, August 08, 2003 2:22 AM To: squid-users Subject: Re: [squid-users] CPU utilization

performance issue

Hi everyone,

thanks for the input. The ACL list have since been slightly altered, using only src(22 entries), dstdomain(114 entries) and url_regex(20 entries). I

am

currently on kernel 2.4.20-19.9 so the

Hyperthreading

might hv been optimized.

Now the machine is handling about 110 req/s but

again

the CPU will climb to abt 90-95%. Is it possible

for

my squid box to go beyond 180 req/s, which is the

peak

for each proxies in the existing pool(ISP env)? I

am

trying to replace my existing NetCaches with squids...one box for one box.

I am wondering if its because reiserfs will consume more CPU than other fs like ext3? Will changing my cache partitions to reiserf lower down the CPU

usage?
Or can anyone suggest other possible improvements?
Thanks.
my 3 caching partitions are on 3 separate disks:-
/dev/sdb1      /cdata1  reiserfs notail,noatime 1 2
/dev/sdc1      /cdata2  reiserfs notail,noatime 1 2
/dev/sdd1      /cdata3  reiserfs notail,noatime 1 2
--
Wolf
--- Tay Teck Wee <[EMAIL PROTECTED]> wrote:

Hi,

when I'm getting about 90 req/s or 800 concurrent connection(according to my foundry L4) to my squid(RedHat 8.0/2.5 Stable3 w deny_info patch),
the

CPU utilization avg abt 80%. How do I lower the

CPU

utilization of my squid? Thanks.

Below is my machine specs:-

Intel Xeon single-processor 2.4GHz(DELL 2650) 2G physical RAM(w 2G swap under linux) 2X 33G for everything except caching (mirror) 3X 33G for caching (volume)

/dev/sda7 505605 68437 411064

15% / /dev/sda1 124427 9454 108549

9% /boot /dev/sdb1 35542688 201248 35341440

1% /cdata1 /dev/sdc1 35542688 200888 35341800

1% /cdata2 /dev/sdd1 35542688 200940 35341748

1% /cdata3 /dev/sda3 1035692 49796 933284

6% /home none 1032588 0 1032588

0% /dev/shm /dev/sda5 1035660 691648 291404

71% /usr /dev/sda6 505605 76236 403265

16% /usr/local /dev/sda8 29695892 83456 28103936

1% /var

Below is my squid.conf(only the essential).

For ACL, basically I hv 3 acl list(in 3 separate files), one containing allowable IPs while the

other

contains deny IPs. I also hv 3 list of banned

sites

list(in 3 separate files).:-

http_port 8080 acl QUERY urlpath_regex cgi-bin \? no_cache deny QUERY cache_mem 400 MB cache_swap_low 92 cache_swap_high 95 maximum_object_size 2 MB maximum_object_size_in_memory 100 KB cache_replacement_policy heap GDSF memory_replacement_policy heap GDSF cache_dir aufs /cdata1 16000 36 256 cache_dir aufs /cdata2 16000 36 256 cache_dir aufs /cdata3 16000 36 256 cache_access_log

/var/log/cachelog/cache.access.log
cache_log /var/log/cachelog/cache.log
cache_store_log none
quick_abort_min -1 KB
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
#3 banned list files
acl SBA dstdomain "/usr/local/squid/etc/SBA.txt"
acl CNB dstdomain "/usr/local/squid/etc/CNB.txt"
acl CNB2 url_regex "/usr/local/squid/etc/CNB2.txt"
#3 access list files
acl NetTP src "/usr/local/squid/etc/NetTPsrc.acl"
acl NetDeny src "/usr/local/squid/etc/deny.acl"
acl NetAllow src "/usr/local/squid/etc/allow.acl"
http_access deny SBA
http_access deny CNB
http_access deny CNB2
http_access deny NetDeny
http_access allow NetAllow
http_access allow NetTP
http_access deny all
http_reply_access allow all
cache_effective_user squid
cache_effective_group squid
logfile_rotate 10
deny_info ERR_SBA_DENIED SBA
deny_info ERR_CNB_DENIED CNB CNB2
memory_pools off
coredump_dir /var/log/cachelog
Thanks again!
Regards,
Wolf
__________________________________________________ Do You Yahoo!? Send free SMS from your PC! http://sg.sms.yahoo.com
__________________________________________________
Do You Yahoo!?
Send free SMS from your PC!
http://sg.sms.yahoo.com
__________________________________________________
Do You Yahoo!?
Send free SMS from your PC!
http://sg.sms.yahoo.com

Re: [squid-users] CPU utilization performance issue

Reply via email to