> -----Original Message----- > From: Carlos Eduardo Gomes Marins > [mailto:carlos.suporte.deinf@xxxxxxxxxx] > Sent: Monday, February 13, 2006 7:55 AM > To: squid-users@xxxxxxxxxxxxxxx > Subject: Help tuning squid > > > Hi, > > I believe I have some performance issues here using Squid. > The swap area is being used a lot (590 MB). > Almost the whole memory (2 GB) is being used, and the Process > Size (via squidclient) is 427710 KB. > CPU is being heavly used - 80% (squidclient). > I use Trendmicro IWSS as default parent. > Here are some data: > Output (part of) of top > > 14:45:07 up 6 days, 1:30, 1 user, load average: 4.01, 3.83, 3.87 > 573 processes: 569 sleeping, 4 running, 0 zombie, 0 stopped > CPU states: cpu user nice system irq softirq > iowait idle > total 58.3% 0.0% 28.3% 0.7% 2.9% > 3.6% 5.8% > cpu00 57.2% 0.0% 25.8% 1.5% 4.9% > 4.5% 5.8% > cpu01 59.5% 0.0% 30.8% 0.0% 0.9% > 2.7% 5.8% > Mem: 2061660k av, 2042204k used, 19456k free, 0k > shrd, 39248k buff > 1063336k actv, 129836k in_d, 30388k in_c > Swap: 5120376k av, 594344k used, 4526032k free > 453808k cached > While there is a lot of swap in use, I would not say that this machine is particularly memory starved. You have over 400MB of memory used as a disk cache (453808k cached) which can effectively be added to the memory buffer (39248k buff) and called "free". The process count is certainly impressive. > PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME > CPU COMMAND > 10986 nobody 25 0 424M 417M 2340 R 40.1 20.7 1055m > 0 squid > 1625 root 15 0 9052 8620 8016 S 2.0 0.4 127:36 > 1 winbindd > 1627 root 15 0 10296 9.9M 9696 S 0.9 0.4 54:48 > 1 winbindd > 2042 iscan 15 0 9976 9976 9284 D 0.8 0.4 21:41 > 1 postmaster > 11373 nobody 15 0 708 676 640 D 0.7 0.0 14:12 > 1 diskd > -------------------------------------------------------------- > ------------------------ > > Data from squidclient: > > PROCESS SIZE > Process Data Segment Size via sbrk(): 427710 KB > Maximum Resident Size: 0 KB > Total space in arena: 427710 KB > > PAGE FAULT RATE > Page faults with physical i/o: 14793 > 5min: page_faults = 0.009999/sec > 60 min: page_faults = 0.035832/sec > > HTTP REQUESTS > Number of HTTP requests received: 7448700 > Average HTTP requests per minute since start: 1766.5 > 5min: client_http.requests = 139.435024/sec > 60min: client_http.requests = 109.345998/sec > > HTTP SERVICE TIME > client_http.all_median_svc_time = 0.047755 seconds Requests are being serviced quickly. > > DNS SERVICE TIME > dns.median_svc_time = 0.170482 seconds DNS is perhaps a little slow. > > OPEN FILE DESCRIPTORS > Number of file desc currently in use: 2145 > > CPU > 5 min cpu_usage = 68.974431% > 60 min cpu_usage = 81.409616% > -------------------------------------------------------------- > ----------------------- > Squid.conf (some performance related stuff) > > maximum_object_size 10240 KB > maximum_object_size_in_memory 20 KB > ipcache_size 10000 > fqdncache_size 10000 > negative_dns_ttl 15 seconds > cache_mem 256 MB > cache_peer 127.0.0.1 parent 8088 7 no-query default > -------------------------------------------------------------- > ----------------------- > I don't have Squidguard nor Dansguardian, only Trendmicro > IWSS as mentioned, so all the acls's and handled by Squid itself. > Is there anything I can do to improve the overall performance? Put the virus scanning on a different box. Make sure your DNS servers are up to the load. Don't use regex acl's. Look into the epoll patch. > Thanks in advance, > > Carlos Eduardo. > > Carlos Eduardo Gomes Marins. > Analista de Segurança em TI. > Cast Informática S/A. > Tel.: +55 (61) 3414.3607 > carlos.suporte.deinf@xxxxxxxxxx > Chris