Conference ssag::ask_ssag

6598.0. "KZPSA and KZPDA performance problems" by FRSAES::SEEMANN () Fri Apr 18 1997 09:36

My Customer "MPI" tried out something.

They have a A4100 and wanted to find out if there is a disk performance
problem with the KZPDA and or a combination of KZPSA and KZPDA.
What they tried is, with UNIX, a DD (64k Blocks) input from a RRZ (raw 
device) out to DEV0 (null device).

KZPDA is at Rev B01.

1. two KZPSA in one PSI segment with 2 9Gb disks connected to each controller

	read from one disk  = 10Mb/s
	read from two disks = 16Mb/s  (one controller only)
	read from two disks = 20Mb/s  (two controllers)
This is what the customer is using as a refference, and finds it to be normal

2. one KZPDA with 2 9Gb disks

	read from one disk  = 10Mb/s
	read from two disks =  8Mb/s  (together only!!!) 
The throughput would be expected to be much higher with two disks but it is 
reduced as can be seen above

3. one KZPDA with 2 9Gb disks and one KZPSA with 2 9Gb disk in one PSI secment

	read from one disk on KZPDA	= 10Mb/s
	read from one disk on KZPSA	= 10Mb/s
	read from two disks,
			 one on KZPDA	= 10Mb/s
			 and one on KZPSA = 3.5Mb/s
	after stopping the reading on the KZPDA
			the throughput on KZPSA immediately jumps
			to 10 Mb/s

These tests were performed with a seagate 9GB disk drive.
It was also tested on RZ29b-VW with the comparable results but with
less throughput.

Further testing on other systems showed comparable results (A600).

First tests were performed with a self written C program under which these
strange results were detected. Then they found out that it could be 
simulated with the simple DD under UNIX.

Any ideas?

Thank you

Hubert