RE: [squid-users] Help tuning squid

> -----Original Message-----
> From: Carlos Eduardo Gomes Marins
> [mailto:carlos.suporte.deinf@bcb.gov.br]
> Sent: Monday, February 13, 2006 7:55 AM
> To: squid-users@squid-cache.org
> Subject: [squid-users] 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@bcb.gov.br
>

Chris
Received on Tue Feb 14 2006 - 18:22:40 MST