Pathway/XM System Management Manual
Overview of Configuring and Managing Pathway/XM
Compaq NonStop™ Pathway/XM System Management Manual—426761-001
3-11
Load Balancing
Requests from Pathsend processes for distributed server classes are handled in the same
way as other transaction requests, with one difference. Pathsend requests are always
directed to the one PATHMON process in the Pathway/XM environment that has an
external name, and are serviced by the server processes controlled by that PATHMON
process.
Load Balancing
To automatically balance the terminal workload across the Pathway/XM environment,
the Pathway/XM software first groups the TERM objects into sets based on their TCP
template. It then calculates the number of terminals that can be supported by a TCP
process, based on the number of server classes and on the TCP template and its
associated NODE specification. This is the number of TCP processes to be started per
set of terminals. The terminals in the set are then distributed among the TCP processes
based on the configured terminal weights, so that all TCPs are assigned the same relative
workload.
Each PROGRAM object is assigned to a single TCP process that has no terminals
assigned. This process is used to create temporary TERM objects when RUN
commands are issued.
Finally, based on the total number of TCP processes needed to support the terminals in a
given Pathway/XM environment, the Pathway/XM software determines the number of
PATHMON processes that must be started. The TCP processes are then distributed
among the PATHMON processes according to the total terminal weight assigned to each
TCP process. In each NODE object that is associated with the PATHMON object and
also has TCPs assigned, one or more PATHMON process pairs are created. Each
PATHMON process is assigned a primary and a backup CPU based on the CPU
distribution specified in the associated NODE object. When each TCP process is
created, it is assigned a primary and a backup CPU based on the NODE specification
assigned to the TCP template.
Pathway/XM also provides additional load balancing features to support Pathway/iTS
web clients.
Availability
Transaction protection for distributed server classes is provided by NonStop™ TM/MP.
If the LCS process fails, it is restarted by its PATHMON process. The requester process
then receives a communication error and should abort or restart the transaction. All
pending transactions sent to the LCS process are lost and must be retried; the server
class is still available, because the PATHMON process has started a new LCS process.
When a processor fails, the server processes running in that processor are no longer
available to handle requests, and the queue length increases. The LCS interprets this
increase in queue length as a need for additional resources. New server processes are
then created on other processors in the logical node, resulting in automatic redistribution
of processes.