New buffer/cache small report and questions

Ricardo Galli Granada (gallir@atlas-iap.es)
Tue, 22 Jun 1999 15:48:24 +0200 (MEST)

I tried for a couple of hours (in a production machine) the 2.3.7 kernel
and tested 100.000 operations (60 % reads, 40 % insert and update) on a
database (Empress). In 2.2.10, the average CPU time per operation was
0.0294 secs, in 2.3.7 was 0.0251 secs.

Furthermore, the sys time in 2.2.20 was about 45% of the total time (55 %
user time), in 2.3.7 is about 38 %. I noticed the difference is
probably due to the table/record locks that are always written to a file
and this could force the double buffer/cache copy everytime a table is
accesed.

Congratulations, the results look very good.

My question is, are the changes going to be back ported to 2.2.x?

My second question is, can a 4096 bytes block size improve results even
more?

Best regards.

PS: the machine was a PII Dual 300 MHz, 512 MB RAM (more than 400 MB of
buffer-cache), with Adaptec 7880 and two 8 GB disks.

--
Ricardo Galli






-
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/