I guess my first question would be, how does the I/O bottleneck manifest
itself?  Are you having poor response time?  Are your processes
significantly paused for disc?  Are you getting messages from a tool like
SOS or Glance reporting I/O bottlenecks?  If you are running a monitoring
tool, are your numbers spiking for brief periods or are they staying
consistently high?

There may be nothing wrong with your system tuning or sizing at all.  It may
be the nature of the application you are running.  For example if you are
doing a serial read of a data base, you may see a rather dramatic increase
in pause for disc.  That's not necessarily a problem that you can solve;
it's the nature of the beast.  Serial reads by definition are always
bringing new chunks of data into memory to be processed.

> -----Original Message-----
> From: Les bureaux de credit du nord [SMTP:[log in to unmask]]
> Sent: Friday, October 30, 1998 10:58 AM
> To:   [log in to unmask]
> Subject:      IO BOTTLENECK ON THE HP957SX
> 
> I experience IO BOTTLENECK on our HP957SX with the following
> configuration:
>  
> 2 C3027U SE SCSI   internal hard disk of the operating system (ldev 1 & 2)
> 5 C2490A FW SCSI disk on first FW io card
> 3  C2490A FW SCSI disk on second FW io card
> Memory capacity 192 Mbytes
> 10 users
> 5.16 Gbytes free of hard disk space
> operating system MPE ix 5.0
> HP30391C.05.08 TurboIMAGE/XL Image prefetch is turned on.
>  
> In an attempt to resolve the problem I have:
>  
> - defragmented the database to create large blocks of data with few
> extents
> - performed a condense of all LDEVS
> - run howmessy software and performed repacks and checked for sysnonym
> chains.
>  
> All my discs seem to be well balanced in terms of IO with a total of 58
> io's /sec.  Most of my database access are keyed access.
> Cpu is utilized to about 30%.   Memory prefetch wait is about 80%.  I will
> be adding 64 Mbytes of RAM next week to further decrease memory pressure. 
> 
>  
> Is my system bus or disk mechanisms too slow and creates an io
> bottleneck?  The batch job I run is written in SPL..  I update about
> 70,000 records/hour.  What can I expect from a 957SX  if cpu was fully
> utilized?  Would the bottleneck result from an incorrectly set system
> table?   Any suggestions you might have would be greatly appreciated.
>  
>  
> Jean Huot
> Northern Credit Bureaus,
> [log in to unmask] <mailto:[log in to unmask]>
> (819) 762-4351
>