Hello everybody!
I run squid 1.1.18 on FreeBSD 2.2.2. It is linked against GNU malloc.
Now my cache swap contains about 700,000 objects, and after squid
restart I get the following mstats :-)
=====
...
        Maximum Resident Size: 187672 KB
        Page faults with physical i/o: 78
Memory usage for squid via mstats():
        Total space in arena:  196928 KB
        Total free:            105803 KB 54%
...
Accounted Memory Usage:
        StoreEntry                 673445 x   52 bytes =  34198 KB
        URL strings                                    =  31425 KB
        IPCacheEntry                   42 x   36 bytes =      1 KB
        FQDNCacheEntry                  0 x   56 bytes =      0 KB
        Hash link                       1 x   12 bytes =      0 KB
        Pool MemObject structures      76 x  100 bytes =      7 KB (     0 free)
        Pool for Request structur      70 x 4536 bytes =    310 KB (     0 free)
        Pool for in-memory object      13 x 4096 bytes =     52 KB (     0 free)
        Pool for disk I/O              23 x 8192 bytes =    184 KB (     0 free)
        NetDB Address Entries           0 x   76 bytes =      0 KB
        NetDB Host Entries              0 x    8 bytes =      0 KB
        NetDB Peer Entries              0 x   24 bytes =      0 KB
        ClientDB Entries                1 x  292 bytes =      0 KB
        Miscellaneous                                  =   1674 KB
        Total Accounted                                =  67853 KB
=====
I was observing the same thing with standard FreeBSD malloc (that
squid's size in memory is twice its accounted size)
It seems that squid's behaviour causes a heavy fragmentation of the arena...
I went doing RTFS...
SY, Yar
Received on Thu Dec 18 1997 - 11:41:47 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:59 MST