[squid-users] High CPU usage problem on Squid 2.6 STABLE9

Hello
i have problem with high cpu usage and browsing latency (now browsing
sites without squid if faster than with squid enabled),

i have 3 SCSI disk in configuration:
cache_dir diskd /cache/cache1 1000 2 256 Q1=64 Q2=72
cache_dir diskd /cache/cache2 1000 2 256 Q1=64 Q2=72
cache_dir diskd /cache/cache3 1000 2 256 Q1=64 Q2=72
and mounted:
/dev/sdd1 on /cache/cache1 type reiserfs
(rw,noexec,nosuid,noatime,nodiratime,notail,nolog,block-allocator=noborder)
/dev/sde1 on /cache/cache2 type reiserfs
(rw,noexec,nosuid,noatime,nodiratime,notail,nolog,block-allocator=noborder)
/dev/sdf1 on /cache/cache3 type reiserfs
(rw,noexec,nosuid,noatime,nodiratime,notail,nolog,block-allocator=noborder)

Server have 2 processors Intel(R) Xeon(TM) CPU 3.06GHz

squid compiled with epool options:
Squid Cache: Version 2.6.STABLE9
configure options: '--enable-storeio=diskd,aufs'
'--enable-removal-policies=heap' '--disable-wccp' '--enable-arp-acl'
'--enable-cache-digests' '--enable-default-err-language=Polish'
'--enable-linux-netfilter' '--disable-ident-lookups'
'--disable-hostname-checks' '--enable-underscores' '--enable-async-io'
'--enable-kill-parent-hack' '--enable-dlmalloc'
'--enable-xmalloc-statistics' '--enable-epoll' '--with-pthreads'
'--disable-poll' '--disable-select' '--disable-kqueue'
and top shows cpu usage between: 45-80%

General Runtime Information:
Squid Object Cache: Version 2.6.STABLE9
Start Time: Sat, 27 Jan 2007 22:25:44 GMT
Current Time: Sun, 28 Jan 2007 12:13:04 GMT

Connection information for squid:
    Number of clients accessing cache: 600
    Number of HTTP requests received: 1207836
    Number of ICP messages received: 0
    Number of ICP messages sent: 0
    Number of queued ICP replies: 0
    Request failure ratio: 0.00
    Average HTTP requests per minute since start: 1459.9
    Average ICP messages per minute since start: 0.0
    Select loop called: 13632428 times, 3.641 ms avg
Cache information for squid:
    Request Hit Ratios: 5min: 27.8%, 60min: 32.4%
    Byte Hit Ratios: 5min: 3.2%, 60min: 7.5%
    Request Memory Hit Ratios: 5min: 26.6%, 60min: 37.1%
    Request Disk Hit Ratios: 5min: 17.4%, 60min: 19.5%
    Storage Swap size: 2764200 KB
    Storage Mem size: 262140 KB
    Mean Object Size: 9.42 KB
    Requests given to unlinkd: 0
Median Service Times (seconds) 5 min 60 min:
    HTTP Requests (All): 0.09736 0.07825
    Cache Misses: 0.15888 0.15048
    Cache Hits: 0.01035 0.00767
    Near Hits: 0.20843 0.08265
    Not-Modified Replies: 0.00919 0.00562
    DNS Lookups: 0.00669 0.00464
    ICP Queries: 0.00000 0.00000
Resource usage for squid:
    UP Time: 49639.941 seconds
    CPU Time: 7749.424 seconds
    CPU Usage: 15.61%
    CPU Usage, 5 minute avg: 68.05%
    CPU Usage, 60 minute avg: 60.43%
    Process Data Segment Size via sbrk(): 380084 KB
    Maximum Resident Size: 0 KB
    Page faults with physical i/o: 0
Memory usage for squid via mallinfo():
    Total space in arena: 380084 KB
    Ordinary blocks: 372509 KB 30299 blks
    Small blocks: 0 KB 0 blks
    Holding blocks: 3744 KB 6 blks
    Free Small blocks: 0 KB
    Free Ordinary blocks: 7574 KB
    Total in use: 376253 KB 98%
    Total free: 7574 KB 2%
    Total size: 383828 KB
Memory accounted for:
    Total accounted: 335021 KB
    memPoolAlloc calls: 147275404
    memPoolFree calls: 145284658
File descriptor usage for squid:
    Maximum number of file descriptors: 8192
    Largest file desc currently in use: 255
    Number of file desc currently in use: 155
    Files queued for open: 0
    Available number of file descriptors: 8037
    Reserved number of file descriptors: 100
    Store Disk files open: 0
    IO loop method: epoll
Internal Data Structures:
    296720 StoreEntries
     62366 StoreEntries with MemObjects
     62303 Hot Object Cache Items
    293437 on-disk objects

This is normal that cpu usage is high or something is wrong ?

Thanks for help
Robert
Received on Sun Jan 28 2007 - 05:17:30 MST