HP-UX System Administrator's Guide: Logical Volume Management (5900-3028, March 2013)
Timeout Differences: 11i v2 and 11i v3
Since native multi-pathing is included in 11i v3 mass storage stack and it is enabled by default,
the LVM timeout concepts may vary between 11i v2 and 11i v3 in certain cases.
• Meaning of PV timeout.
In 11i v2, LVM utilizes the configured PV timeout fully for a particular PV link to which it is
set. If there is any I/O failure, LVM will retry the I/O on a next available PV link to the same
physical volume with the new PV timeout budget.
But in 11i v3, when native multi-pathing is enabled, the mass storage stack does the I/O
retries on all/subset of the available PV links within the single configured PV timeout period.
In spite of this difference, the default PV timeout is retained as 30 seconds in 11i v3 to
guarantee backward compatibility.
Care should be taken in selecting an appropriate value for the PV timeout in 11i v3, so that
retries may be attempted on all available PV links within the timeout period.
• Timeout considerations for mirrored LV environment.
With 11i v3, in certain situations, LVM I/Os to a mirrored Logical volume may take longer
time to complete (comparing to 11i v2) than the configured LVM LV timeout. It happens when
one or more PV links goes offline while write I/Os to a mirrored logical volume are in-progress.
The situation gets worse when DasProvider, SAM or lvlnboot runs while PV link goes offline.
However this situation remains only for some period of time (usually up to transient_secs
but could be higher in some cases), until the associated LUNs transitions out of transient state
either to online or offline state, after which it is recovered automatically. This could be due to
the higher timeout value set in the SCSI tunable transient_secs and path_fail_secs.
If the physical volume has few links underneath, reducing these parameters may result in faster
I/O completion.
transient_secs and path_fail_secs are mass storage stack tunable parameters that
can be configured through the scsimgr and scsimgr_esdisk commands. Regardless of
these parameter settings, the mass storage stack never holds onto any physical I/O originating
from LVM for more than the PV timeout budget. These two parameter settings never replace
the PV timeout but control the threshold of a LUN and Lunpath before failing I/Os.
CAUTION: Lowering the parameters transient_secs and path_fail_secs may result in less I/O
resiliency.
For more details on these two tunable parameters, see the scsimgr(1M) and scsimgr_esdisk(7)
manpages. Also refer to the Path Recovery Policies and Managing Loss of Accessibility to a
LUN sections of the HP-UX 11i v3 Native Multi-Pathing for Mass Storage whitepaper on http://
www.hp.com/go/hpux-core-docs.
NOTE: When you use disks with a single link, such as most internal disks, with 11i v3 you
can set SCSI parameters transient_secs and path_fail_secs to the PV timeout value
to get similar 11i v2 timeout behavior in 11iv3 for the mirrored LV I/Os.
164 LVM I/O Timeout Parameters