On Tue, 5 Sep 1995 13:00:00 -0500 paul (p.) shields said:
>You should acheive maximum performance and uptime by mirroring two sets of
>stripped volumes. You get all the write benefits of stripping and all the
>read benefits of mirroring. Plus you never have any down time. When a volume
>fails you put a new one in and resynch the databases.

>This is the configuration being recommended by IBM. The problems we are

Well, I can see where that could give you good performance.   Going all
out say two seperate RAID disk arrays with decent disk space, each w/4 2GB
disks, you begin to talk some serious money though.  I guess just for
kicks you could add a dual controller to each array as well.  What would
all that cost ~$50K in IBM equipment?  Mucho dinero there.

I would guess a poor man's solution to performance could be to spread the
database over two plain FW/SCSI-2 disk arrays and mirror them, use some
1GB sub-7ms read access disks to spread the database over the maximum
number of spindles.  Since the 2GB drives are about 8ms you'd gain 12-15%
increase for the average read but, I think the average write is the same
about 9ms.  Even this solution would probably cost what, ~$25K.

Price/performance issues can be such a pain sometimes.

>having is that our backups have a very small average file size, thus require
>a lot of database activity and overhead during backup. What i want to really
>know is with an average file size of 20-40K what kind of performance are
>other sites seeing? If performance is breaking 1500Kb/s, how is you system
>configured to acheive this? We are seeing performance in the 500-1000 Kb/s
>range and this is very dissapointing.

---
Keith A. Crabb         Keith AT UH DOT EDU
Keith A. Crabb         Keith AT UH DOT EDU
University of Houston  Operating Systems Specialist +1-713-743-1530