Proactive Host Failover - IBM Storwize V5010 Manual

4. Click Create. A window opens indicating that the task of setting up the throttle completed,
as shown in Figure 5-120.
Figure 5-120 IOPS throttle set
5. Click Close.
The following considerations have to be taken into account when defining the I/O throttle on
host or Host Clusters:
I/O throttle cannot be defined for host if it is a part of Host Cluster which already has an I/O
throttle defined at the Host Cluster level.
If the Host Cluster does not have an I/O throttle defined, its member hosts can have their
individual I/O throttles defined.
The mdiskgrp (storage pool) throttles for child pool and parent pool work independently.
If a volume has multiple copies then throttling would be done for the mdiskgrp (storage
pool) serving the primary copy. The throttling will not be applicable for the secondary pool
for mirrored volumes and stretched cluster implementations.
A host cannot be added to a Host Cluster if both of them have their individual throttles
defined.
A seeding host used for creating a Host Cluster cannot have a host throttle defined for it.

5.7 Proactive Host Failover

During planned maintenance procedures of IBM Storwize V5000 Gen2, such as firmware
upgrade or node canister replacement, at times the host multipathing driver may experience
issues recovering paths to volumes. Most of the issues seem to stem from the fact that the
path to a volume was removed without warning to the host and the host has to detect that the
path is now offline, time-out, and re-drive all I/Os that were in flight.
As a secondary issue, hosts often attempt to restart sending I/O to the preferred node
canister as soon as the ports come online but the node canister may still be unpending.
During this time, the node canister queues all incoming I/O until it has its configuration data
and then volumes start coming online.
Chapter 5. Host configuration
285