Bug in buffer cache?

Peer Griebel (griebel@uni-paderborn.de)
Wed, 6 Sep 1995 11:22:59 +0200 (MET DST)


in the last days I analyzed the speedup of using the md-package (raid-0). I
therefore patched the iozone program to do random reading and writing. (The
following tests don't have anything to do with md. md was just the reason of
the tests. The tests were perfomed on a 486/33 16MB ram, aha1542, 650MB+1GB

I observed that writing randomly 512 byte blocks on 512 byte boundaries
results in heavy traffic to the harddisk although all the writing should be
done in buffer cache.

The test itself creates a 10MB file and afterwards writes 10MB randomly to
the file. A reading phase of 10MB is done after that.

I have 16MB ram so the whole file should be kept in core.

Why do I get this result???

If I change the block size to 4096 bytes (the block boundary is still 512
bytes) the result is as expected. All writing and reading is done in core.
the harddisk remains quiet.

Strange...?! Can somebody explain that?

connection reset by

Dipl.Inform. Universitaet-GH Paderborn dying is easy
Peer Griebel Fachbereich 17 - Informatik it's living
Tel.: 05251-60-2076 D-33095 Paderborn that scares me
griebel@uni-paderborn.de Raum C2.305 (A. Lennox)
<A HREF="http://www.uni-paderborn.de/fachbereich/AG/szwillus/griebel.html">
P. Griebel</A>.