N, see, Balancing disk processes – HP NonStop G-Series User Manual
Page 149
Balancing and Tuning a System
Measure User’s Guide — 520560-003
7- 13
Checking and Tuning Problem Areas
Too many index levels in a key-sequenced file, poorly written applications, and the
wrong cache size for disk processes all generate unnecessary disk I/Os. Use FUP
to check and correct the index levels in key-sequenced files. For more information,
see
.
3. Distribute I/O activity evenly across all the disks. Because of application limitations
and the volume of I/O activity, balancing disk activity can be a complex operation.
These steps describe one approach to the problem:
a. Balance swap activity across the disks. To examine swap activity, list the disks
by their SWAPS counter values:
16+ LIST DISC *, BY SWAPS
To balance swapping, move object files off a heavily used disk or specify
SWAP FILE in the RUN command or the call to NEWPROCESS. Having
multiple copies of object files can aggravate swapping because code pages
are not shared. This is especially true if you create multiple versions of heavily
used object files such as library and system files. A more efficient practice is to
divide copies of object files onto multiple volumes. For example, place objects
A through M on one volume and objects N through Z on a second volume.
b. Balance disk queues, as described in
Balancing Disk Processes
Examine your system diagram. Distribute primary I/O control processes evenly across
all CPUs. The I/O control processes are all high-priority processes. If you concentrate
them in one CPU, they contend with each other for CPU time.
If most of the I/O processes are on a few CPUs, you probably will have to reconfigure
the system and run SYSGEN on the system again. However, if the I/O processes are
distributed evenly, you might be able to balance the load by moving the primary I/O
process to a different CPU.
If your system diagram shows a number of heavily used disks on a single controller,
the disks are likely contending for use of the controller, which can cause a performance
problem.
The workload should be about the same on all PINs except the last. The last PIN
should be operating at about 5 to 10 percent of capacity.
Checking I/O Activity
A single logical I/O operation on a key-sequenced disk file typically requires from one
to three physical I/O operations. To determine the ratio of physical to logical I/O
operations for a disk, add the READS and WRITES counters for the disk (physical
Note.
On newer disk controllers, the SEEK-BUSY-TIME will be zero. Seek times are not
provided. To calculate estimated time per I/O request:
DISC-BUSY / (READ-RATE + WRITE-RATE)