HP3000-L Archives

February 2001, Week 3

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Mark Landin <[log in to unmask]>
Reply To:
Mark Landin <[log in to unmask]>
Date:
Wed, 21 Feb 2001 16:31:30 -0600
Content-Type:
text/plain
Parts/Attachments:
text/plain (54 lines)
On Wed, 21 Feb 2001 13:15:25 -0600 (Central Standard Time), Michael
Anderson <[log in to unmask]> wrote:

>We are trying to come up with the right hardware configuration to purchase a refurbished 979/ks400. This investment is something that we plan on having around for a long time, after all it is an HP-e3000. It will definitely out live all the NT boxes around here. So I want to make sure that I get exactly what is necessary for the long term, 5-10 years.

If you can wait a little while, the new A-class MPE boxes will get you
more bang for the buck, most likely.

>I would like to hear from others on the list who have been there, done this. Mistakes made, regrets, advice on reliability and performance. Specifically in the DISC I/O area, but info in other areas will also be appreciated. We would like to spec this new box out with the fastest disc I/O possible. Reliability is most important. Disc I/O as always been a problem with our applications here. We would prefer multiple fiber channels/controllers with a hot swop disc array, RAID or something like that. Plus a separate I/O channel to the internal system drive.

Unfortunately, 9xx boxes have only really supported two viable disk
I/O flavors: FWD SCSI (20 Mbytes/sec) and FibreChannel (100
Mbytes/sec). Further, HP has never really made a truly screaming disk
array for the 3000 market. (OK, the XP48/256/512 might be a screamer,
but it's price tag is sobering ...)

For arrays, your choices from HP are Model 10, Model 20, and AutoRAID
12H (FWD SCSI), and Model 30/FC, and XP48/256/512 (FibreChannel).
Maybe the FC/60...? Except for the XP arrays, which again are
eye-poppingly expensive, all these products are fairly mature and none
would be considered exceptionally performant.

Honestly, the best way to optimize disk I/O performance is to avoid
disk I/O as much as possible. Max out the memory on whatever box you
are buying, AND on each array disk controller. Use 3rd-party indexing
products on your IMAGE databases to avoid serial reads. Use "fast I/O"
products like Suprtool for read-only report applications. Spread I/O
loads across as many channels, controllers, and disk spindles as
possible. For RAID arrays, hardware RAID controllers are faster than
MirrorDisk/iX. RAID level 0/1 gives better performance than RAID
levels 3 or 5, although you need a lot more disk devices to get the
same amount of usable disk storage for 0/1. Use private volumes so
that the XM manager for your databases don't all thrash LDEV 1. Avoid
CM KSAM files in favor of NM KSAM files. Write new applications with
avoiding disk I/O in mind. Consider using static subroutines instead
of dynamic ones so you don't have to reinitialize static data
structures every time you call the sub.

>Questions:
>What about the core I/O on the box, should it have the same number of I/O channels as the disc array has controllers?

Core I/O only has one SCSI channel, which services all the internal
disks and tape drives you might have. You almost certainly would be
better off getting an additional SCSI I/O card PER disk array
controller that you plan to purchase. (One array can have 2
controllers .. another good performance idea).


>I've heard the term 'I/O channels', and the term 'disc controllers'. Are these terms really referring to the same thing?

No. Multiple disk controllers can be run off of the same I/O channel.
However, for best performance, one disk controller per channel is
advised.

ATOM RSS1 RSS2