STAT.NLM Release Notes - Updated 8/9/93


DOCUMENTATION

   Documentation for STAT.NLM and DUMPSTAT.EXE is a
   NetWare Application Note entitled "STAT.NLM: A Tool for
   Measuring NetWare v3.11 Server Resource Utilization," March
   1992. The information in this Release Note updates that found
   in the AppNote.


DUMPSTAT.EXE OUTPUT FILE CONTENTS

   During the DUMPSTAT process, two variables recorded by
   STAT.NLM--Number of Polling Loops and Maximum Number of
   Polling Loops--are removed from the data and replaced with a
   CPU %Utilization figure. Therefore, the output file includes
   seven rather than eight columns with the following headings:

   CPU %Utilization
   Bytes Received
   Bytes Transmitted
   Bytes Read
   Bytes Written
   Packets Routed
   Number of Connections


DPATCH.NLM

   DPATCH.NLM patches the NetWare v3.x disk process. This
   patch enables NetWare to supply the disk statistics required by
   STAT.NLM's Bytes Read and Bytes Written columns. Always
   load DPATCH.NLM before loading STAT.NLM.


STAT.NLM TRIGGER ANOMALIES

   If you are using the start and stop trigger mechanism to define
   a test duration, you must define a test duration that does not
   include the current time. For example, if the desired test
   duration is from 7:30a to 5:30p, then you must manually
   configure the triggers some time after 5:30p and before 7:30a.

   Triggers set on the hour are often ignored by STAT.NLM.


API TO NETWARE'S INTERNAL STATISTICS VARIABLES

   NetWare's internal statistics can be exported for use by
   application programmers. For more information, see "NetWare
   v3.x Operating System Statistics Exposed!," NetWare
   Application Notes, July 1991.


CPU %UTILIZATION ACCURACY

   The algorithm used by NetWare to calculate CPU %Utilization
   includes a subtle caveat (See the Appnote for further detail). The
   algorithm uses the Maximum Number of Polling Loops statistic
   as the denominator, so it is the reference point from which the
   CPU's utilization is calculated. However, the Maximum Number
   of Polling Loops statistic is only accurate after the CPU idles for
   one second while STAT is active.

   In most production environments, both MONITOR.NLM and
   STAT.NLM provide accurate CPU measurements because the
   CPU regularly rests within 5% of a complete idle state. When
   MONITOR.NLM is pre-loaded as part of the server's
   AUTOEXEC.NCF, the Maximum Number of Polling Loops
   statistic has many opportunities to become accurate before
   active users log into the network. 

   In some environments however, especially benchmarks,
   STAT.NLM can provide inaccurate samples. For example, if the
   CPU is running steadily at 50% utilization before STAT.NLM is
   activated, a brief STAT.NLM measurement during the
   benchmark will be inaccurate because the CPU never hits an idle
   state and the Maximum Number of Polling Loops statistic will
   be approximately 50% of its potential. A work-around in this
   scenario is to activate STAT.NLM before the benchmark is
   started. This sequence of events allows STAT.NLM to capture an
   accurate Maximum Number of Polling Loops statistic before the
   benchmark is begun. Afterward, the CPU %Utilization
   calculations will be accurate.


FUTURE RELEASES

   We have no plans to release another version of STAT.NLM.