[THIN] RE Subject: RMS

  • From: Frith Brown <frith.brown@xxxxxxxxxxxxxx>
  • To: "'thin@xxxxxxxxxxxxxxx'" <thin@xxxxxxxxxxxxxxx>
  • Date: Sun, 30 Jun 2002 10:23:21 +1200

"Fred" wrote......
From: "seskjohor" <seskjohor@xxxxxxxxx>
Subject: RMS
The Thread Count on our Citrix Server in RMS is flagging RED and I 
was wondering what is happening and if it is something that needs 
looking after etc, etc, etc.
Please help.
[Frith]  On a Citrix server, "interrupts/sec" (or threads) can reach quite high
values due to the input (via ICA) from client-side input devices, including
mouse and keyboard activity.  To illustrate this, you can artificially generate
interrupts by quickly moving a mouse back and forwards, or clicking repeatedly,
from an ICA session.
[Frith]  So what is normal behaviour for "interrupts/sec"?  Well, like most
processor counters, "interrupts/sec" will normally sit at a fairly regular
baseline value and spike intermittently.  Quite high spikes for all processor
counters are well-tolerated by Microsoft operating systems, but where spikes are
sustained there are limits beyond which performance will suffer.  In Citrix
systems, sustained user load can result in baseline rising for quite prolonged
periods of time, and the real test is whether users report poor performance. 
[Frith]  But what is a "high value" for "interrupts/sec"?  OK, the reason this
is a difficult question is that "interrupts/sec" is an absolute counter (i.e. it
is not relative to the speed of the processor), and faster processors (and in
particular, later generation processors) can tolerate and continue to
performance under much higher "interrupts/sec" than was previously possible.
These extracts from TechNet illustrate what is considered tolerable, for various
processors, and also shows a trend toward higher "interrupts/sec" being
considered acceptable:
"On an Intel 486 or later processor, the processor clock interrupts every 10
milliseconds, or 100 times per second. Network interrupts can produce 200-1000
"On a Pentium computer, 500 interrupts per second may be normal. A high level
can be as many as 1,000 interrupts per second."
"This counter should operate continuously between 100 and 1,000, but spikes up
to 2,000 are acceptable."
"However, if the number of interrupts consistently exceeds 1,000 on a
80486/66-based system, or 3,500 on a Pentium 90 PCI bus system, a hardware error
or interrupt conflict with devices may be occurring. "
"Depends on processor. Up to 1,000 for 486/66 processors; 3,500 for P90; more
than 7,000 for P200. "
[Frith]  So what level of "interrupts/sec" should I consider problematic, and
configure alerts for?  In my professional opinion, the TechNet extracts above
are illustrative of an increasing acceptance of higher "interrupts/sec",
provided that the processor is of later-generation, high-speed capacity.  This
changing trend is the reason that monitoring utilities such as Resource
Management Services frequently ship with unnaturally low alert lines for
"interrupts/sec":  the utility cannot know what processor the system is running
and the counter is an absolute value.  In the case of a Pentium III processor in
excess of 500 MHz, "interrupts/sec" should not be considered problematic until a
value of at least 7,000 is sustained for at least 60 seconds.  This
consideration is supported by the latest references in TechNet:
"Select the Processor: \Interrupts/sec object:counter, and configure an alert to
be generated when the value is over 7,000 for 60 seconds."
[Frith]  I hope this information is helpful to you.
Kind regards,
Frith Brown
Engineering Services Manager 
Computerland New Zealand


Other related posts:

  • » [THIN] RE Subject: RMS