Wednesday, May 18, 2005

Performance Testing

Tried the dd with block sizes of up to 4096k and there was not much
difference.

e.g.

time dd bs=64k if=/disk1/abigfile of=/dev/null
10002+1 records in
10002+1 records out
real 2m52.13s
user 0m0.04s
sys 0m14.87s

time dd bs=2068k if=/oraback100n0/backup/DB1DV/RMAN/HO>
309+1 records in
309+1 records out
real 3m48.01s
user 0m0.00s
sys 0m8.20s

And the performance makefiles is a bundle of fun. Have a look at these:

time mkfile 100m grunt
real 0m22.89s
user 0m0.03s
sys 0m4.00s

time mkfile 650M grunt

real 3m17.82s
user 0m0.21s
sys 0m25.64s

With the results of 4MB/s it's good enough to say that something is
wrong :) It would be better if you add 'bs=1048576' to your 'dd' command.
And, if you can afford it, test the device (/dev/dsk/c?t?d?s?) instead
of the filesystem.

You might try running 'iostat -xtcn 1' when you do the test, to see how
fast the devices are really accessed at a given time.

0 Comments:

Post a Comment

<< Home