Pathway/XM System Management Manual

Managing Servers
Compaq NonStop™ Pathway/XM System Management Manual—426761-001
9-11
Queue-Management Algorithm Used by the LCS
Processes
To restart a stopped server process, you must first thaw the server process, then issue the
START SERVER command, as shown in the following example:
>> THAW SERVER *
>> START SERVER *
For the syntax of the commands used in stopping a server class, refer to the following:
FREEZE SERVER
on page 11-24
STOP SERVER on page 11-96
THAW SERVER on page 11-106
START SERVER on page 11-59
Queue-Management Algorithm Used by the LCS
Processes
Each LCS process maintains the following statistics related to the current performance
of the transaction queue for its distributed server class:
•
Average transaction arrival time
•
Average transaction service time (S-time)
The LCS process starts one or more new server processes when the arrival rate is greater
than the service rate. A server process is dropped when its reservation has expired. The
number of server process reservations will never become greater than that specified by
the MAXSERVERS attribute for the server class or less than the number specified by
NUMSTATIC for the server class. The number of server processes may at times be
greater than the MAXSERVERS value, when a reservation has expired and the server is
currently processing a request. The server process will be terminated when the request
has completed.
CPU assignments for new server processes are selected in a round-robin fashion
according to the weights specified in the configuration of the NODE object associated
with the server class.
If the CMON attribute is set to ON for the server process, when a CPU is selected the
selection is sent to the $CMON process, if one is present in the system, for approval or
modification of the selection.
When a CPU fails, it is dropped from the list of available CPUs in the general
configuration. The weight assigned to the failed CPU is spread across the remaining
available CPUs such that the total weight for all available CPUs still equals 100%.
When a failed CPU is reloaded, the CPU is again added to the list of available CPUs,
The weight values for each available CPU are restored to their original configured
values, and any remaining failed CPUs once again have their weight values distributed
across the remaining available CPUs. The number of server processes per CPU are
adjusted by stopping processes as their reservations expire.