How many HBAs are you using (and what type) ? If it's just one, you may be out of gas on the interrupt CPU. 50K IOPS translates to 20us of interrupt service per IO. Maybe the interrupt service path got a tad longer ? BTW, Solaris 10 does not account for interrupt service time correctly (i.e. show a high %sys in mpstat), you'll need to measure it with dtrace or lockstat.
50k IOPS is not something most people would try on a single HBA. While it is technically possible using tiny blocks, an iosize of 8KB would have the transport (2Gbps) throttle the iorate to 25K IOPS. In a realistic situation, for sequential IO, you would use a larger block size and less IOPS. For random IO, say 8KB, you would plan to not exceed 250 IOPS/disk, so that would be a configuration of 100 disks on a single HBA. Rocky This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
