HP SAN Virtualization Services Platform 3.0.5 Release Notes (5697-1031, June 2011)

• Any DPM failure will cause a DPM reboot. Previously, some failures would not cause an expected
reboot, and leave the DPM in an uncertain state, requiring a manual reboot.
• Improvements were made to the resource recovery algorithms for synchronous mirrors to provide
improved stability and better behavior under high-stress workloads. Previously, a high-stress
workload could trigger a significant degradation in synchronous mirror write performance, as
well as poor wire-level behavior. This could manifest as reduced performance at the hosts, or with
fabric-level errors (for example, dropped frames, delayed frame delivery, and fabric congestion).
• The counter logic was fixed to handle an internal elapsed time counter rollover, which occurred
after a DPM had been running uninterrupted for 497 days. Previously, a bug in the port sequencing
logic could cause individual ports on the DPM to stop processing incoming commands after the
time counter rollover.
• Additional diagnostic information has been added to the DPM save_state package to enhance
the debugging or wire-level error conditions encountered while in operation.
• Fixed an issue in the resume logic, so all paused commands are successfully resumed and
completed. Previously, it was possible for an operation under heavy load to be paused (to prevent
other operations from being starved of attention), but never resumed. This would cause an I/O
to never complete, which eventually would result in the host timing out the command, aborting it,
and retrying. Operations affected by this issue would eventually complete upon retry, but their
latency would be dramatically increased, affecting overall performance.
• Large synchronous mirror write operations can now be paused and resumed later when
appropriate. Previously, long running operations that occupied a DPM port processor had to be
paused to prevent starvation of other operations. However, certain cases, like large synchronous
mirror write operations, were not paused. This could lead to fabric congestion, performance
problems, or link events on heavily loaded configurations.
• Removed excessive offending log messages that resulted when a DPM does not properly handle
split I/Os, and generates a large number of errors in the DPM logs. In a configuration with a
large number of such errors, the log mechanism could be overwhelmed, leading to performance
or behavioral problems, and masking the condition that actually triggers the errors.
Prerequisites
A dual-fabric SAN as defined by the HP SAN Design Reference Guide (http://www.hp.com/go/
SANDesignGuide), with approved storage arrays using B- or C-series switches. See the HP Single
Point of Connectivity Knowledge (SPOCK) website at http://www.hp.com/storage/spock for supported
switch firmware. Site registration is required.
Installation instructions
For a new or upgrade installation of Virtualization Services Manager or Data Path Module software,
contact HP Support.
If you are installing the VSM GUI, change the registry as described below and reboot the VSM server.
At the location HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\CrashControl\
CrashDumpEnabled, set the registry key to 0x1 (instead of 0x2).
Host operating systems and multipath drivers
See http://www.hp.com/storage/spock for a listing of supported operating systems and required
multipath drivers. Site registration is required. Also see “Operating system issues” (page 11) for issues
with supported operating systems.
To obtain the HP MPIO Full Feature DSM and DSM Manager:
1. Go to http://www.hp.com and click Support & Drivers.
2. In the Step 1 area, select the Download drivers and software (and firmware) option. The Step 2
field becomes available.
4 Prerequisites