HP-UX 11i v3 Mass Storage I/O Performance Improvements

Figure 2 shows I/O throughput results on a system with a configuration similar to that in Figure 1,
using legacy DSFs which are all associated with a single HBA port. On 11i v2, or with load-
balancing disabled, all the I/O goes through the one HBA port. On 11i v3, with load-balancing
enabled, the I/O gets balanced automatically across paths through all four HBA ports, producing in
this case about 3.6 times the throughput of the load-balancing disabled case. Careful static balancing
of LUN path usage in an application or workload can produce similar results, but requires much more
configuration and storage management work and is not guaranteed to continue to provide optimal
results across configuration or workload changes.
Figure 2 I/O throughput (four paths per LUN)
0
200
400
600
800
1,000
1,200
1,400
1,600
HP-UX 11i v2
412 410
HP-UX 11i v3
1,470 1,465
Reads Writes
Throughput - MB/s
(10^6 bytes/sec)
I/Os performed on a set of legacy DSFs, all through a single HBA port.
I/Os performed on the same set of legacy DSFs. Native Multi-Pathing
automatically takes advantage of paths through the other 3 HBA ports.
3.6x 3.6x
11i v2
11i v3 11i v3
11i v2
Figure 3 shows I/O operations per second (IOPS) results on a system with 2 paths (through 2 HBA
ports) per LUN, resulting in a doubling of the performance with load-balancing enabled as compared
to HP-UX 11i v2. These results show slightly more than 2x improvement due to cache locality and
other tuning in the mass storage stack and the kernel in HP-UX 11i v3.
4