HP NonStop G-Series - Manual (Page 146)

Balancing and Tuning a System
Measure User’s Guide — 520560-003
7- 10
Checking and Tuning Problem Areas
Checking and Tuning Problem Areas
This subsection explains how to check and tune or balance common problem areas
found in the system. When attempting to tune a system, start with the highest-level or
most global external change. If the system problem is not resolved, progress to more
complex or lower-level tuning tasks. This strategy is, in order of priority:
1. Balance memory consumption and minimize swapping.
2. Balance disk activity.
3. Balance CPU activity, which includes:
•
Balance other I/O process activity.
•
Balance user process activity.
Step 1: Balance Memory Consumption and Minimize
Swapping
A page fault occurs when a process requires a page not currently in memory. In
response, the memory manager brings the required page into memory, causing one or
two swaps. A high swap rate in a CPU indicates a problem. Possible causes include:
•
The CPU does not have enough memory.
•
The CPU has too many primary disks primaried in the CPU.
•
The disks have too much cache configured.
•
Too many pages are locked into memory.
•
There are too many process creations or deletions.
Applications also can cause excessive swapping by repetitively allocating and
deallocating extended data segments in their program logic.
Swapping causes performance problems for two reasons:
•
Excessive swapping causes unnecessary disk I/O operations.
•
The CPU time required to swap pages in and out of memory could be spent
performing more useful work.
A value greater than 2 swaps per second in the SWAPS counter of the CPU entity
indicates a possible problem. To determine the impact of swapping on your system,
use this formula to determine the maximum cost of swapping in terms of disk I/Os:
No. swaps in all CPUs * 2 = No. disk I/Os for swapping
For example, on systems running D-series RVUs, a disk can reasonably handle 20 to
35 I/Os per second. If five CPUs each generate two faults per second, the result is 20
I/Os per second. Thus, the system is devoting much of a disk to swapping, causing an
unacceptably high fault rate.