[page cache newbie == yes]
I downloaded dbench and have been running it on a 2.3.48 4-way
200 Mhz with 256 Mb mem with ips SCSI adapter (IBM ServerRaid).
I created a command file that does cat /proc/meminfo once every
second, so I can monitor cache usage/free mem. I wanted to get
an idea of the working set of dbench as far the page cache goes.
After I reboot the system, I have :
MemFree: 229024 kB
Cached: 10984 kB
OK, looks reasonable.
If I run dbench 2 one time, I see "Cached:" peak at about 61000 kB during
the run and then drop back to 15196 kB when dbench completes.
Throughput is reported at about 27 MB/sec.
So it looks like 22.5 MB per dbench client.
If I run dbench 5, I see "Cached:" peak at about 135000 kB during
the run and then drop back to 19408 kB when dbench completes.
Throughput is reported as 15 MB/sec.
Ok.
If I run dbench 10 and wait until "MemFree:" is < 10 MB, and then,
[this may be the important part] kill dbench 10 with <ctrl-c>,
"Cached:" does not drop back.
"MemFree:" stays at about 6000 kB.
The next run of dbench 2 completes with a throughput of 3.6 MB/sec,
an 85% decrease.
Hmmmmm.
After the (3.6 MB) dbench 2 completed, "MemFree:" returned to about
50000 kB.
I ran dbench 2 again with only a throughput of 4.81 MB/sec reported.
Now, if I delete the files that were left over by the <ctrl-c> of dbench
10, then "Cached:" drops back to 7336 kB.
If I again run dbench 2, then throughput recovers to 28 MB/sec.
dbench 2 and dbench 10 are running on different physical disks in
the system. The file system in EXT2. The SCSI is ips (IBM ServerRaid).
Is something is not right here ? Why aren't pages being kicked
out of the cache and being reused if I <ctrl-c> ?
Bill Hartner
IBM Linux Technology Center
bhartner@us.ibm.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:23 EST