A couple of things to consider - Linux does not write
to the device. It writes to the page space, then the
page manager, driven by bdflush, writes to the device.
So you would not expect as high a performance under
Linux as you would for MVS.

I/O is backed up behind a single subchannel to a
single device, so the maximum you would expect is "17
mb/sec". Personnally, I have never seen anything over
15 mb/sec with assembly level programming and BSAM. If
you have multiple data sets open on a single device,
then the queue time (wait for the I/O to actually
start) can be large, artificially slowing the I/O
(that's what PAV's amelioriate).

In Linux, the best solution seems to be to use RAID0
for striping or LVM for striping, allowing multiple
subchannels to be active at the same time.

As to shark, the speed to a single raid array is much
less that you would expect - 40-60 mb/sec, even if the
box is rated at 150-300 mb/sec. To get maximum
throughput, the stripes should be over as many raid
arrays (LCU's) as possible.

=====
Jim Sibley
Implementor of Linux on zSeries in the beautiful Silicon Valley

"Computer are useless.They can only give answers." Pablo Picasso

__________________________________
Do you Yahoo!?
New Yahoo! Photos - easier uploading and sharing.
http://photos.yahoo.com/

Reply via email to