dean.sun@gmail.com (Dean) wrote in message
news:<a2efdbbd.0411112105.d210392@posting.google.c om>
....
> I did a write IO test in IBM P690 and HP SUPERDOME.
> When I tested the local disks in P690,use this:
> I used the same method to test the Storage of HDS 9980,which is
> connected to the P690 and SUPERDOME.
> -----------------------------
> the result of P690 is 56MB/s almost,and result of SUPERDOME is 90MB/s
> almost !!!
>
> I'll cry for explaining this to the customer.
> Anyone can help me to explain this?


Maybe You start thinking about it beforce crying?

at first
- unless You plan to be swapping a lot, 56MB/s would still be ok.
- maybe You should consider filesystem caching as a reason? It's
benefit will disappear with 10 Oracle databases running
If You meant to test disk throughput use
vgdisplay -v /dev/vg00 | grep /dev/dsk # get disk disk below /tmp/
dd if=/dev/rdsk/<harddisk> of=/dev/null bs=1024k # look how fast it is
I guess You know how to do the same on AIX?
- I'm quite sure the cpu time used in /dev/null is quite negletible on
a p690, especially for something like 100MB/s - we might talk about
that on a 200MHz sgi doing >800MB/s.
- has it ever occured to You the the superdome might simply have been
shipped with different disks? (if hp used 73GB 15K drives in the
surestore 2100, 100MB/s would be a fair guess)
- Did You look into filesystem mount parameters? (e.g. mounting on
hp-ux with 'nodatainlog' would surely be a point to search at for the
difference in performance)

- also both 56 and 100MB/s are just fast, not really very fast, unless
the Hitachi is highly loaded. For external storage I'd expect
something better, so
some baseline tuning might be reasonable.
- check the fc adapters - maybe it's something trivial like a broken
gbic in
the p690, also check if it's 1 or 2Gbit FC Ports they are connected
to.

Two general points:

The Superdome IMHO has a bad ratio between cpu's and I/o. Ask HP how
many PCI-X slots You get when using the usual configuration of 4 I/O
drawers. If I remember correctly it's only 4 per drawer. Which is not
so much for 128 CPUs.
If Your customer really needs high I/O potential, this might not be
the perfect choice, and the p690 will probably much better suited.

To some of the other posters:
I agree in that dd really is not a reasonable benchmark. But first
things come first! Horrible performance on /tmp? (and 7MB/s is far
from what one should expect today, not even 4 years ago). I wouldn't
care for the difference in between 30 and 40 MB/s, but to get down to
7MB/s something has to be wrong.
In addition to that, if it were a memory filesystem and only made
100MB/s it would be even worse
I get 3-7MB/s on my stone-age 4.3GB seagate disks in a B50 at home.
both the p690 and the superdome a bit heavier on the wallet, and it's
just
fair Dean expects a bit more performance.

Regards,
Florian