OS/2 FAQ List: User's Edition (4 Apr 94)

(5.7) Measuring Performance and Memory Usage

How do I measure OS/2 performance and memory usage?

OS/2 does not treat system resources like DOS.  Memory is treated as a 
virtual resource, used intelligently.  For example, OS/2 will retain 
unused, "dormant" code in memory if that memory is not otherwise required, 
on the assumption that that code may be used again.  Also, all but a small 
portion of OS/2 (and most applications, no matter how many are running) 
may be paged to disk should a large amount of physical memory be required. 
 Utilities which display "free" memory, then, are only useful for rough, 
relative measurements.  (Such utilities also often fail for another 
reason: many only report the largest contiguous block of free physical 
RAM.  And a few will never report more than 16 MB of RAM because they were 
designed for OS/2 1.x.)

Similarly, utilities which purport to measure system load (e.g. Pulse) 
should not be relied upon for definitive performance measurement. 
 Subjective assessments are often much more reliable.  Pulse (and similar 
utilities) rely on a measurement of processor time allocated to a thread 
running at OS/2's lowest priority.  This method is sometimes subject to 
erroneous results.

That said, more rigorous system performance optimization and monitoring 
tools include System Performance Monitor/2 (IBM Program No. 5871-3415), 
BenchTech (Synetik, phone 303-241-1718), OR/SysMon (International OS/2 
User Group, phone 44-285-641175 or FAX 44-285-640181), CPU Monitor (Bon 
Ami), and Performance 2.1 (Clear & Simple, phone 203-658-1204).

Note that OS/2's swap file is designed to behave with hysteresis.  It will 
not shrink in size as easily as it grows, under the assumption that swap 
space needed once may be needed again.  It should shrink given enough time 
and continued, less intense system loads.

Related information:

(5.6) Performance Tuning

Inf-HTML End Run - Successful