Working with queuing network models

A queuing network model (QNM) estimates service performance against changes in business driver volumes or infrastructure configuration. It can predict:

  • Response time
  • Throughput
  • Queue length
  • Utilization

These metrics are either computed for a single resource or for a complex network; predicted values are then automatically compared with the service levels specified as thresholds on response time and utilization.

The following table specifies the system resource types supported by QN models, including the corresponding metrics:

Model resource type

Metric label

Metric name

CPU

CPU_UTIL

CPU utilization %

Memory

MEM_UTIL

Memory utilization %

Disk

DISK_UTIL

Disk utilization %

Input/output bus

CPU_UTIL_WAIO

CPU % time waiting for I/O

The model wizard will enforce these constraints by preventing you from selecting unsupported metrics.

For more information about how to work with E-QNM, see the following topics:

Example of the QN model type

The following figure shows an example of how an input-output table for a QN model appears in BMC Helix Capacity Optimization.

Example of an input-output table for the QN model

The following table describes the Business Drivers (formerly known and shown in the figure as Workloads) and Systems in the example.

Business drivers

Systems

  • The model contains two business drivers: number of visits and orders received.
  • Based on the current (metric) value that is available, a target value can be set that determines the cutoff after which the model estimation terminates.
  • Target value is determined by target growth factor, which is the ratio between target and current value.
  • The current setup will not support a target growth factor of 4 because the upper limit predicted by the model is 2.05.
  • BMC Helix Capacity Optimization warns you that the target values exceed the model thresholds by displaying the affected values in red.
  • A red indicator in the far-right column of the table alerts you that the given business driver violates the response time of 5 seconds.
  • The model contains a set of 6 systems.
  • The multiplicity of every existing system or resource can be chosen, and its speed variation be calculated based on the built-in benchmark data.
  • If the current number of CPUs are increased for the Apache_cluster system from 8 to 12, the calculated utilization of 33% does not violate the maximal permissible utilization of 70% (green indicator).
  • If, for the system bea_frontend, the number of CPUs is decreased from 8 to 5, a red indicator appears, showing that the utilization will be exceeded.
  • To obtain a better response time, a what-if analysis can be performed.

The following figure shows the response time chart for the example.

Response time plot before computer upgrade

Beyond the given threshold, the response time experiences an exponential increase in value (the region shaded red).

The current system can be improved, as shown in the following figure.

Upgrading the computers to see the effect on the response time

Applying this change yields the plot shown in the following figure.

Response time plot after computer upgrade

The response time drops compared to the earlier values.

With BMC Helix Capacity Optimization, you can simulate various scenarios to arrive at the solution that best corresponds to your needs.

Was this page helpful? Yes No Submitting... Thank you

Comments