Tandem Capacity Model (TCM) Manual

17. Disk Seconds—Modifiable through the Change Transaction option of the Update tab
Disk Seconds is the actual and planned consumption, or cost, per transaction in disk seconds
for each transaction.
18. Avg Resp Time—Not directly modifiable
Avg Resp Time is the host response time for each transaction and the average or specified
percentile transaction. This number includes both CPU and disk response times. It does not
include communication line time and therefore is not end-to-end. Expected OLTP response
times for CPUs are calculated using a standard M/M/1 with priority queueing model. Expected
OLTP response times for disks are calculated with the M/U/1 model. Total OLTP host response
time in a system is the sum of the CPU response time and disk response time. The actual
average response time, as well as the plan, are estimates based on queueing theory and are
not measured directly.
Batch transaction response time is calculated using a multiple priority closed queueing model
and is always the average response time. Specified percentile response times apply only to
OLTP. For more information, see “Modifying Percentile Response Times” (page 90), Chapter 7:
“Modeling Batch Workloads”, and Appendix D: “TCM Response Time Formulas”.
Waited writes of any TXN are either all audited or all not audited. If audited, TMF Audit trails
will give the number of audit trails required. If not audited, writes are either serial or parallel
on all mirrored disks.
The demand numbers used to compute host response time do not include the demand for fixed
categories, it is assumed that transactions do not wait for fixed activity to complete. However,
fixed consumption is included in the utilization figures.
The average transaction response time is the expected host response time for the (weighted)
average OLTP transaction or the unweighted average batch transaction. The value in the Plan
Average Response Time field for the average transaction must be less than or equal to the
number in the Plan Average Response Time target field described previously, if one has been
entered. For individual transaction types, these fields reflect the expected host response time
for that type of actual or planned transaction.
Disk response time depends on the disk configuration type (audited, nonaudited, unmirrored,
or serial). For response time formulas that TCM uses, see Appendix D: “TCM Response Time
Formulas”.
Planning Timeline Model Screen
The Planning Timeline model is used to examine the impact of various upgrade scenarios and is
integral to the capacity planning process. It examines the changes in CPU and disk utilization
levels and host response times for eight time periods as throughputs, numbers of CPUs and disks,
and types of CPU and disk configurations vary. This model is useful for capacity planning when
specific changes to hardware configuration, transaction throughputs, or both are anticipated or
proposed.
Each Planning Timeline model shows eight scenarios. The first column shows the Actual sample
data from the Consumption model. The user can select either the first plan scenario data (second
column) from either the Actual data or the Plan data of the Consumption model. The default shows
the Actual Data. The remaining scenarios are all related to the planned scenario in the second
column. Each throughput level is incrementally higher than the one prior by the percentage specified
in the Rate column. Figure 46 (page 145) screen.
144 The Performance Model