-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hi all:
I'm doing some comparison studies on dasd / disk I/O for different
configurations of storage and storage on different virtualization
platforms (trying to convince management that oracle on Z is an okay
thing). I'd really really appreciate any advice regarding how to
structure and go about doing this testing.
What I'm thinking about is testing the following:
*) our dual, controller based mirrored DS4800s providing EKCD DASD
*) Some FCP storage used as EKCD minidisk via z/VM (forget the proper
term here)
*) Direct guest attached FCP luns
Our FCP luns would be coming from a set of HP XP controllers.
Lacking other ideas, I plan to use a suitably intensive set of IOzone
test runs as my benchmark - trying to test both large and small I/O
operations.
I'd plan to do all this testing on an up to date install of rhel 5.4 in
a single IFL guest, tuned to minimal memory (just barely swapping) with
a vdisk high priority swap and an EKCD low priority swap.
All the testing would be done on a Z9 LPAR with 2 IFLs, 22 GB of primary
storage, 2GB of extended storage, and running z/VM 5.3
*) For EKCD DASD:
a single minidisk with no other linux guests on it, configured as the
only PV in a volume group, 1 stripe, with one LV, using ext3 with
default settings. Served from an IBM 4800 disk controller doing
controller based mirroring
z/VM Minidisk PAV + dm tunedasd
cache multipath cache size
|=============|===============|===============|============
| EXCD | yes | yes | 50
|=============|===============|===============|============
| EXCD | yes | yes | 10
|=============|===============|===============|============
| EXCD | yes | yes | 0
|=============|===============|===============|============
| EXCD | yes | no | 50
|=============|===============|===============|============
| EXCD | yes | no | 10
|=============|===============|===============|============
| EXCD | yes | no | 0
|=============|===============|===============|============
| EXCD | no | yes | 50
|=============|===============|===============|============
| EXCD | no | yes | 10
|=============|===============|===============|============
| EXCD | no | yes | 0
|=============|===============|===============|============
| EXCD | no | no | 50
|=============|===============|===============|============
| EXCD | no | no | 10
|=============|===============|===============|============
| EXCD | no | no | 0
|=============|===============|===============|============
| EXCD/FCP | yes | yes | 50
|=============|===============|===============|============
| EXCD/FCP | yes | yes | 10
|=============|===============|===============|============
| EXCD/FCP | yes | yes | 0
|=============|===============|===============|============
| EXCD/FCP | yes | no | 50
|=============|===============|===============|============
| EXCD/FCP | yes | no | 10
|=============|===============|===============|============
| EXCD/FCP | yes | no | 0
|=============|===============|===============|============
| EXCD/FCP | no | yes | 50
|=============|===============|===============|============
| EXCD/FCP | no | yes | 10
|=============|===============|===============|============
| EXCD/FCP | no | yes | 0
|=============|===============|===============|============
| EXCD/FCP | no | no | 50
|=============|===============|===============|============
| EXCD/FCP | no | no | 10
|=============|===============|===============|============
| EXCD/FCP | no | no | 0
|=============|===============|===============|============
*) For direct guest attached FCP/SAN storage:
a single LUN configured using NPIV, using round robin dm-multipathing
with 2 paths (note: pending our storage team's commentary), configured
as the only PV in a volume group, 1 stripe, with one LV, using ext3
default settings:
Since I don't have the option to have z/VM minidisk cache here, I'm
going to mess around with the amount of memory linux is allocated, to
see how well it behaves with more or less memory available for buffer cache.
We don't do controller based mirroring on our XP's, so I'd like to test
host based mirroring to an identical lun, also.
Note the OS itself would still be installed on EKCD dasd.
linux mirroring
memory |none |via lvm|via mdadm
|==============|=======|=======|=========
| minimal | yes | no | no
|==============|=======|=======|=========
| minimal | no | yes | yes
|==============|=======|=======|=========
| minimal | no | no | yes
|==============|=======|=======|=========
| min+256k | yes | no | no
|==============|=======|=======|=========
| min+256k | no | yes | yes
|==============|=======|=======|=========
| min+256k | no | no | yes
|==============|=======|=======|=========
| min+512k | yes | no | no
|==============|=======|=======|=========
| min+512k | no | yes | yes
|==============|=======|=======|=========
| min+512k | no | no | yes
|==============|=======|=======|=========
I'd really appreciate any advice as to how to go about this. Anything
ranging from 'you're silly to even consider doing this in light of X, Y,
and Z' to 'here's how I did something similar' to 'this benchmark would
be much better'. Or anything else at all, really.
Thanks for your time!
- -- Pat
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
iEYEARECAAYFAktsshIACgkQNObCqA8uBsxMaQCeNi0p/RJNQuciCHrWUzE4TNwo
2ocAoJG5q7OWkawf7Mkx5kJpcUXoJ95K
=w5Ok
-----END PGP SIGNATURE-----
----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
