Executing A Failover - HP e3000 MPE/iX Manual

If any other error type occurs (such as a data transmission or device error), the
I/O subsystem manages the error and perform corrective action. HAFO remains
idle and not participate. In addition, HAFO remains idle when error types are
received from non-high availability devices.
The difference between a Hung device and an extremely slow device can be
CAUTION
very difficult to determine. It is possible that under extreme I/O loads a device
may become so slow in responding that a HAFO event is triggered. Before
adding HAFO to an existing system, Disk I/O subsystem performance should be
analyzed for possible performance bottlenecks, especially around "peak"
processing times. (Refer to Chapter 4, sections "Configuration Planning" and
"ADDCONF" for more information.)

Executing a Failover

When a trigger status is received, HAFO immediately begins the failover
sequence. This sequence activates the alternate data path and reroutes I/O to it.
Failover occurs on a per Ldev basis. Each device manager (the piece of the I/O
subsystem that manages a specific Ldev) detects a data path failure
independently as a result of an I/O to its Ldev. If the I/O rate is low, it is
possible that Ldevs on a failed path will not learn of a failure for a noticeable
period of time however, as I/O failures occur, each Ldev will failover. For
example, if three Ldevs on a fast-wide SCSI bus experience an array controller
failure, the associated three device managers will perform a HAFO failover
event independently. Figure 2-2 illustrates a failover event in a sample system.
Figure 2- 2
Ldev 1
Ldev 30
File
System
Ldev 31
Volume
Mgmt
Ldev 101
Ldev 102
Ldev 103
MPE/iX
DM
DM
DM
DM
DM
DM
Failover
High Availability Array
Logical Physical
(view)
D
D
A
SCSI
A
bus
M FC
D
D
A
A
SCSI
M bus
FC
Product Description
Components
A
1
r
r
a
30
y
C
o
n
31
t
A 101
r
r
a
y
102
C
o
n
103
t
11