HSJ80 supported disks.

[John Welsh_2]

Hi all,
I need to upgrade some disks on a HSJ80 running V8.5J-2 firmware. My aim is to replace 36GB 10K disks with 15K disks. The Release Notes for V8.5J show only 10K 36GB disks.
Is there a later version of firmware that will support 36GB 15K disks ? I have searched for newer versions of ACS and Release Notes but haven't been successful. Thanks, John W.

[Bob Blunt]

John, at this point "supported" is a somewhat moot point for the HSJ80. There are later firmware versions than V85J-2 (V85J-7 and -8)but I think the updates were for break-fix rather than new device support.

V85J-10 was the last release of ACS for the HSJ80 on PCMCIA cards but not many of my customers ever got it. I had loaded the patches in our lab to the V85J-2 PCMCIA cards and the releases thru V85J-7 basically filled the available patch area on the controller. The release notes for V85J-10 and -11 (which requires the -10 cards) say nothing about higher speed disk support.

I hate to say it but you're probably on your own for the 15K disks. Did you get them from HP as replacements for 36GB/10K drives?

bob

[John Welsh_2]

Thanks Bob.
I have located a Quickspecs dated 26-Aug-03 for the HSJ80/EMA8000 that shows one 36GB 15K disk in the list of available disks for F/W version V8.5J-2.
So it looks like we can use those.
We are trying to improve I/O for a VAX7000 cluster and need to squeeze some more oomph from the HSJ80. Regards, John W.

[Bob Blunt]

More oomph from HSJ80s, huh? AND we're talking CI... There are a lot of areas to examine there. The latest ACS won't hurt but the latest CIXCD firmware would be better, I think 49? I'll have to check past emails, there were some new changes to "enhance" CI thruput that did NOT work as well as planned.

It would be interesting to see what the stats on your CI are, as well as a look at the console output from the HSJs and errorlog data, SHOW CLUSTER output, etc. Single or multiple CI rails, how many HSJs, what types and how many hosts?

The 15K disks can help some, but if the CI isn't performing like clockwork the effort can be wasted.

bob

[Hein van den Heuvel]

The drives are nearly, but admittedly not entirely, transparent to the controller. The scsi backend controller will negotiated maximum connection speed as appropriate. I expect pretty much any disk to 'just work' specially if as you indicated some 15K drive was officially supported. I would make sure they do not use (much) more power than the current drives. Check the 5V and 12V Amps if listed on the physical drive stickers.

The 15Kpm have two potential benefits.
1) Reduced rotational delays which are a significant component of the total seek time, That's always good.
2) Increased MB per seconds as the bit fly by faster. That's unlikely to help much as it will be limited by that scsi backend controller, and even more so by the severly limited CI maximum throughput. So it may help in response time for a RAID-5 controller configuration by making the backend quicker, but don't expect more MB/sec towards the VAX.

>> We are trying to improve I/O for a VAX7000 cluster and need to squeeze some more oomph from the HSJ80. Regards, John W.

As per above, if responce/seek time is the goal, then there is good hope. If the Oomph needed is in the MB/second space, then I expect no improvement.

The best way to speed up an IO is to make it go away. Has everything possible be done to tune say RMS or RDB to minimize the required IOs?

Typically one would like to tune for larger buffers / pages / buckets. But in the limited CI bandwidth case you may want to minimize those as long as that does not introduce too many more IOs (make sure not to add an index level for example).

I suppose the application/OS is locked in time/platform/version?

Cheers,
Hein van den Heuvel
HvdH Performance Consulting

[John Reagan]

Besides the power requirements as Hein mentioned, also monitor the increased heat from the 15K drives. From my experience the lack of support for newer drives sometimes is due to fans and heat not from the drives themselves.

[John Welsh_2]

Thanks guys,
We plan to install 15K drives and then do some VMS and Database tuning and also spread out the heavily accessed disks across the HSJ's ( 2 x dual HSJ50's and a dual HSJ80
with two CIXCD's per VAX7000 and two star couplers). I am still investigating the possibility of running 4K packets but it looks like support was withdrawn for 4K packets on the CIXCD's. Anyone have input on the 4k packet support ?
I will close this off shortly and assign points. Best Regards, JohnW.

[Bob Blunt]

John, I'm curious about the configuration of your J80s. Are they installed in an individual controller shelf connected to standard StorageWorks device shelves with capacity for six disks? Installed in a modular controller shelf with individual shelves holding up to 14 drives each? Installed in a BA370 (integral controller shelf with integral 4x6 device bays) with either one or two BA370s cross-connected together?

In either configuration I'd spend some time watching the J80 output of VTDPY to see how the load on your J80s is working out and try to balance things as much as possible. I haven't, personally, setup a J80 with the modular shelves with up to 14 hot swap drives per shelf nor do I know if it was ever officially supported.

[John Welsh_2]

Thanks guys for your assistance.
The customer is delaying making any
changes for the present even though
his performance isn't exactly ideal.
I will close this for now.
Thanks, John Welsh.