High Availability by Walter McCullough

but if there is only one pathway to the data then it
will queue I/O requests for
the drive mechanism and wait for each I/O completion until proceeding.
The next problem with this configuration is a
major reduction in throughput
caused by MPE/iX's use of the
Transaction Manager. This mechanism keeps
track of changes to
MPE/iX internal data structures and user data bases, such
as KSAM and Image/SQL files. The Transaction Manager's (XM's) log
file lives
on the master volume of a volume set. If there is
one volume set then all activity
will be to that volume set's master.
As the many applications did their random I/O
to their data files, the Symmetrix
internal cache was of no
use because of the lack of good hit rate. The
customer's random access I/O characteristics showed a hit rate of only
30%. A
cache hit rate of 75% must be achieved before any I/O
performance gains are
realized when four 1Gb stand-alone disk
drives are consolidated into one 4Gb
Symmetrix drive.
"Rules of Thumb"
For Improving I/O Performance and Migrating Data to New Disk Drive Technologies
1. Limit the disk size or partition size from
1Gb to 2Gb and have several pathways
(LDEVs) for the system volume set. MPE/iX uses opened files on disk as extensions to
its memory and does "file operations" through its memory
manager. This operating
system works best when there are
multiple pathways to the working set (transient
space) and permanent data.
2. Restore the application's data to User
Volumes. Create as many User Volume Sets
as make sense. This reduces the XM bottleneck to the master volume of the volume
set, adds more pathways to the data and adds fault
containment. This will limit the
amount of data to reload in case of catastrophic disk failure.
3. Gather data on the user's application I/O
characteristics. There are several tools
available like GLANCE/XL that can help collect this data.
4. Do not exceed 10-20 I/O operations per
second per physical device. Consolidating a
large number of physical devices to a fewer number of larger capacity devices
might
cause you to exceed the recommended I/O rate and cause
an I/O bottleneck. For
better performance throughput it is
also recommended that NO MORE THAN 4
DRIVES BE CONNECTED TO
THE SINGLE ENDED BUS and 8 FOR FastWide.
(Your mileage will vary depending on driving habits)
5. Purchase a high availability disk drive
technology, like the Nike Array or the
Symmetrix 3000, for its feature set and its ability to protect data. Performance of
these
products is dependent on a clear understanding of how it
is used, I/O characteristics of
the application, and the way it is configured.
» Return to original page
Privacy statement Using this site means you accept its terms Feedback to webmaster
© 2008 Hewlett-Packard Development Company, L.P.
Page
4
of
4
High Availability disk products for the HP e3000
7/18/2008
http://www.hp.com/cgi
-
bin/pf
-
new.cgi?IN=http://jazz.external.hp.com/papers/ha.html