The Question is:
When running an application such as SAS, which uses asynchronous Direct I/O, on
an AlphaStation 500au, against locally attached UltraSCSI disks (5 disks,
software RAID 0, FWD SCSI bus), read/write performance is poor compared with
other similar OS's and d
oes not respond to changes in the blocksize being transferred. The CPU
generally sees about 50% utilization, the disks are observably in intermittent
use (access lights flash). The SAS job is the only process on the system.
1) is it possible that the Storageworks RAID driver "packetizes" or otherwise
generates multiple small synchronous I/O's, rather than single large
2) are there other tuning or hardware parameters which might result in low-rate
synchronous I/O on this system?
The Answer is :
Unfortunately, a specific answer will require detailed knowledge of
the application and particularly of the application I/O activity and
I/O processing. Potential areas where performance could be improved
involve the application file I/O, process quotas, system parameter
settings, highwater marking, file extension processing, I/O (read
and/or write) caching, available physical memory, I/O caching
settings, and any number of other areas. Application profiling
is often required, and the assistance of OpenVMS Engineering and
particularly the assistance of the softwarae vendor can be central
in an expeditious resolution.