We are running in a data warehouse environment using extensive 11g RDBMS partiioning and parallel query servers on fairly good sized tables. All of the storage is ASM based with external RAID 0+1 redundancy configured on SUN StorageTEK SAN ( fibre channel HBA's unswitched) The physical rAID device consists of only 8 10K RPM spindles
Oracle enterprise Manager's ASM performance page is telling me that the cumulative stats for I/O response time are 75.79 ms and that average throughput is about 1/MB per second. This seems pretty slow and consistently my query performance reports are showing extensive I/O waits.
What are other people seeing for aberage response time /I/O throughput for read intensive well tuned systems?
(because we are using parallel query the PX coordinators are using the large pool rather than the buffer cache so the majority of the I/O is direct reads rather than scattered I/O)
My question is would increasing the number of spindles in the existing ASM disk group significantly reduce I/O response time and increase throughput.
With an expansion tray I could go up to 16 spiondles -- could I comfortable assert to my management that such an outlay would halve the ASM I/O response time and double the I/O throughput -- i highly doubt it will scale that literally but would like to know if increasing spindle count would definitely reduc e the I/O latency we are witnessing....
Avg resp Time (ms) Throughout (MB/second)