High Availability Storage Options and Their Impact on Performance

8
The XP512/XP48 is no longer available from HP but may be still available in the used product market.
HP StorageWorks Disk Array XP1024/XP128
The XP1024 (and smaller XP128) are HP's latest entry into 2Gbit per second fibre channel connectivity.
The XP1024/XP128 again supports multi-host heterogeneous environments and along with HP's latest
(last) released version of MPE/iX 7.5 on PCI based hardware support native fibre channel (no router) to
the portfolio of HP FC switches and FC hardware.
For those customers still using older NIO (HVD-SCSI only) based hardware can upgrade to the latest
version of MPE/iX and use the latest version A5814A-003 SCSI-Fibre Channel Fabric Router. This newer
hardware supports the newer VA family and XP family of storage along with the portfolio of FC
switches. (The A5814A is not an upgradeable product to the A5814A-003.)
The XP1024 (and XP128) support an even larger cache along with previous features like Continuous
Access XP and Business Copy XP. MPE/iX version 7.0 and 7.5 support an enhanced version of MPE's
Failover/iX (I/O path failover) and also the Cluster/iX solution for even greater availability.
As the storage technology has improved over time in both performance and connectivity, customers have
increased the number of servers connected and this has increased the demands of that storage. This
increase in demand has sometimes impacted performance on individual servers like the HP e3000.
The increase in performance of the disks and their ability to transfer data to and from the array cache and
controllers has helped reduce the impact of cache misses. One of the largest impacts to the overall
performance of the array is the effective reduction of the amount of array cache available for each of the
servers. As the number of servers (and workload) increased, the amount of array cache available to (just)
one server decreased (like to the HP e3000). Changes in the workload of other the servers also has been
known to drastically change the performance of jobs and applications running on the HP e3000 (and other
servers) that are sensitive to disk I/O traffic delays.
Another performance issue observed which impacts performance on the array is MPE/iX XM checkpoint
operations. The checkpoints have a lesser impact on overall performance when a large amount of array
cache is available to handle the operation. In the vast majority of instances this is not the case and these
checkpoint operations do tend to saturate the array cache which can impact both the performance of the
HP e3000 and in some cases other servers connected to the array.
Depending on the application and system I/O activity, the frequency of the checkpoints can become
(somewhat) predictable. MPE/iX can be re-configured to reduce the frequency of these events. When the
XM log file size is increased the frequency of checkpoints decreases but the interesting observation is that
the duration of each of those checkpoints do not significantly increase.