The following DBH load parameters can be used to control processor utilization:
PP DEACTPP RESULT-DELAYPP SCHEDULINGPP SERVERTASKPP WAIT
The load parameter PP DEACT
The load parameter DEACT can be used to control how server tasks are to be handled by the operating system in high-load situations.
If DEACT=NO is set, server tasks are not affected much by other tasks in the system and thus operate more efficiently even in high-load situations where a high percentage of system resources are used and in wait states. This could, however, have an adverse effect on other applications.
The load parameter PP RESULT-DELAY
The load parameter PP RESULT-DELAY can be used to group request results for user tasks and thus improve throughput in high-load situations.
Note, however, that if PP SCHEDULING=ASYMMETRIC has also been set, the number of request results actually grouped may be restricted.
The load parameter PP SCHEDULING
The load parameter PP SCHEDULING can be used to control the internal behavior of multiple server tasks in peak-load situations with a high degree of fluctuation.
PP SCHEDULING=SYMMETRIC causes all server tasks to process requests equally. It is useful for well-utilized server tasks or in cases where no high BS2000 priority can be assigned to the server tasks.
This setting is the default value, since the best performance in cases where loads fluctuate dynamically is obtained by having all requests processed equally by the server tasks, i.e. by a setting appropriate for the median load.
PP SCHEDULING=ASYMMETRIC causes the load on the server tasks to be unequal. This allows the processor cache to be used more efficiently and also enables better coordination with the task scheduling of the operating system, especially in cases when the server tasks cannot be fully utilized, but where even one server task less would be insufficient.
The load parameter PP SERVERTASK
In order to utilize the processors efficiently, it is essential that the load parameter PP SERVERTASK be set correctly. Appropriate consideration must be given to BS2000 priorities to achieve the optimum effect for the setting (see the section “Performanceoriented BS2000 settings” for details).
Due to the multi-threading facility, a single server task (PP SERVERTASK=1) is usually sufficient on a monoprocessor when using asynchronous I/O (PP IO=ASYNC). The use of additional server tasks is only advantageous on multiprocessor systems. The upper limit for the load parameter PP SERVERTASK is usually (i.e. with PP IO=ASYNC) the number of processors.
Note, however, that a higher value is also accepted under certain circumstances, i.e. if the load parameter PP IO=SYNC is specified.
The appropriate setting for the load parameter PP SERVERTASK can be determined for a normal situation (PP IO=ASYNC) with a high user load by means of a simple measurement. This is done by setting the load parameter PP SERVERTASK to the possible upper limit in the measurement configuration and then determining the CPU utilization of the server tasks, which can be detected from the program name UDSSUB. The measurement for CPU utilization can be performed easily at the start and end of a measurement phase by using the BS2000 command SHOW-USER-STATUS.
If you add the CPU utilization of tasks with the program UDSSUB (detectable with SHOW-USER-STATUS INFORMATION=PROGRAM) and then set this value in relation to the CPU utilization of all active tasks in the system, you will obtain a good approximation of the proportion of CPU resources taken up by server tasks in each measurement phase.
In order to determine the “correct” value for the load parameter PP SERVERTASK, you can then multiply the maximum value across all measurement phases with the number of processors on the system.
For example, if the server tasks collectively use 41% of CPU resources on a quadprocessor system, you will need 1.64 (=0.41 x 4) server tasks. The appropriate setting would then be PP SERVERTASK=2.
If a configuration is subject to a high fluctuation in load, the setting for PP SERVERTASK can be adjusted to suit the requirements for the peak load (i.e. the best response times) or an average load (i.e. optimum CPU utilization during normal operation).
Note that setting the load parameter PP SERVERTASK too high will increase task communication significantly, i.e. result in higher CPU utilization and thus possibly impede other applications.
A setting that is too low, by contrast, will generally reduce the throughput rate in the UDS/SQL configuration.
The load parameter PP WAIT
The load parameter PP WAIT can be used to set an active waiting strategy.
Under normal circumstances, if the load is low, waiting for new requests to arrive is controlled via eventing. This ensures that the CPU can also be used by other applications if no further requests can be processed by the server tasks.
If it is possible to ensure through the application that such “underload” situations are extremely rare or very brief, the CPU overhead required for eventing in the normal and peak-load case can be reduced by active waiting.
PP RESULT-DELAY and PP WAIT are intended for peak-load configurations and should only be used if the conditions described for these load parameters are actually fulfilled (see RESULT-DELAY in chapter "DBH load parameters" and WAIT in chapter "DBH load parameters"). Otherwise, they could have a strong adverse effect on operations.