Slowness of Squid without any apparent reason

                                
Our largest cache in France is an AlphaServer 1000 (Digital Unix 4.0)
running Squid 1.0.18. It is incredibly slow, taking several seconds to
deliver a very short HTML page (already in the cache) and sometimes
dozens of seconds. The variation of response times is important. (A
neighbor on the same regional network, with E1 links at 2Mb/s, serves a
file in less than a second.)

During tests, the machine had a normal behaviour and speed. Once we start
to register the clients, it became a snail.

Configuration :

cache_mem 130
cache_swap 14500
        
(We have 320 Mbytes of RAM. The cache disk is 100 % full.)

ps shows:

USER PID %CPU %MEM VSZ RSS TT S STARTED TIME
COMMAND
root 7256 8.1 73.8 314M 235M ?? R Oct 16 09:30:19
/usr/local/squid/bin/squid

(Notice the memory is much more than the size of memory-cached objects.)

The disks are Digital RZ29 SCSI, accessed through a RAID 0 controller
StorageWorks 200. The network attachment is a FDDI DAS.

The machine has 26 clients and one neighbour. It receives two TCP
connection per second (and three UDP in the same time).

monitor shows a small swapping (pout = 6), 600 kb/s on the cache disk, 80
packets/s on the FDDI link. Load is between 0.1 and 0.2.

Here is the output from cache_object://localhost/info:

Getting cache_object://localhost/info from proxy cache.cnrs.fr, port 3128
{Squid Object Cache: Version 1.0.18}
{Start Time: Wednesday, 16-Oct-96 14:11:22 GMT}
{Current Time: Friday, 18-Oct-96 13:50:57 GMT}
{Connection information for squid:}
{ Number of TCP connections: 327078}
{ Number of UDP connections: 465327}
{ Connections per hour: 16626.3}
{Cache information for squid:}
{ Storage Swap size: 13571 MB}
{ Storage Mem size: 54891 KB}
{ Storage Expiration at: Thursday, 01-Jan-70 00:00:00 GMT}
{Resource usage for squid:}
{ CPU Time: 34257 seconds (27677 user 6580 sys)}
{ CPU Usage: 20%}
{ Process Size: rss 1944000 KB}
{ Page faults with physical i/o: 365766}
{Memory usage for squid via mallinfo():}
{ Total space in arena: 290079 KB}
{ Ordinary blocks: 86915 KB 636482 blks}
{ Small blocks: 146070 KB 1671700 blks}
{ Holding blocks: 0 KB 16717 blks}
{ Free Small blocks: 4063 KB}
{ Free Ordinary blocks: 53029 KB}
{ Total in use: 232986 KB 80%}
{ Total free: 57092 KB 20%}
{File descriptor usage for squid:}
{ Max number of file desc available: 4096}
{ Largest file desc currently in use: 878}
{ Available number of file descriptors: 3452}
{ Reserved number of file descriptors: 100}
{ Active file descriptors:}
{ ( 0 = LOG) STDIN}
{ ( 1 = LOG) STDOUT}
{ ( 2 = LOG) STDERR}
{ ( 3 = LOG) /var/log/squid/cache.log}
{ ( 4 = -1, 0) NET /usr/local/squid/bin/dnsserver #1}
{ ( 5 = -1, 0) NET /usr/local/squid/bin/dnsserver #2}
{ ( 6 = -1, 0) NET /usr/local/squid/bin/dnsserver #3}
{ ( 7 = -1, 0) NET /usr/local/squid/bin/dnsserver #4}
{ ( 8 = -1, 0) NET /usr/local/squid/bin/dnsserver #5}
{ ( 9 = FILE) /var/log/squid/hierarchy.log}
{ ( 10 = FILE) /var/log/squid/access.log}
{ ( 11 = PIPE) ftpget -S}
{ ( 12 = PIPE) ftpget -S}
{ ( 13 = FILE) /var/log/squid/store.log}
[...]
{Stop List:}
{ HTTP:}
{ /cgi-bin/}
{ .cgi}
{ ?}
{ GOPHER:}
{ ?}
{Internal Data Structures:}
{ Hot Object Cache Items 468}
{ StoreEntries with MemObjects 468}
{Meta Data:}
{ StoreEntry 798256 x 64 bytes = 49891 KB}
{ IPCacheEntry 924 x 88 bytes = 79 KB}
{ Hash link 37 x 24 bytes = 0 KB}
{ URL strings = 37591 KB}
{ Pool MemObject structures 512 x 152 bytes = 76 KB ( 6
free)}
{ Pool for Request structur 512 x 4384 bytes = 2192 KB ( 205
free)}
{ Pool for in-memory object 19200 x 4096 bytes = 76800 KB ( 21340
free)}
{ Pool for disk I/O 200 x 8192 bytes = 1600 KB ( 704
free)}
{ Miscellaneous = 4898 KB}
{ Total Accounted = 173129 KB}
}

Stephane Bortzmeyer Institut Pasteur
bortzmeyer@pasteur.fr Service d'Informatique Scientifique
                              Paris, France
+33 01 40 61 34 62

http://web.cnam.fr/personnes/bortzmeyer/home_page.dom
Received on Fri Oct 18 1996 - 06:57:39 MDT