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

The load-balancing policies operate on the list of available paths to a LUN, which can change as
paths go offline or online due to error conditions or by a user disabling paths with the scsimgr
command. For more information, see scsimgr(1M) and
The performance benefits of load-balancing include the following:
Increased I/O throughput through balanced utilization of all available paths.
Decreased CPU utilization (and increased aggregate throughput capability) on cell-based machines
using cell-local load-balancing.
Increased I/O throughput
The use of load balancing can significantly increase the I/O throughput capability of a system. For
example, Figure 1 shows a system with four HBA ports, each of which is connected to a set of LUNs
in a set of storage arrays.
Figure 1. Sample system with four HBA ports
Storage
Arrays
HBA
port 1
HBA
port 2
HBA
port 3
Storage
Arrays
HBA
port 4
SAN
HP-UX Serve
r
Figure 1 – System with 4 paths per LUN
Without load-balancing, I/O requests might be overloaded on some paths while other paths are
under-utilized. Load-balancing spreads the load evenly across the available paths to make use of the
full bandwidth of the various paths and avoid degradations associated with overloading a specific
path. This can be especially important in high workload situations or where consistent service levels
are required.
HP testing shows significant throughput and I/O operations per second (IOPS) increases with load-
balancing enabled (which it is by default in HP-UX 11i v3). These tests were performed using the Disk
Bench I/O benchmark tool on an rx6600 server with 4 Gb/s AB379B Fibre Channel adapters
connected to MSA1500 Fibre Channel disks. For comparison purposes, the same set of legacy DSFs
was used in all these tests. For further comparison the tests were performed using the corresponding
new persistent DSFs, without any change in the 11i v3 results. The actual I/O performance increase
experienced on a customer system will depend on the system workload and the mass storage
hardware and software configuration.
3