2.4 The Database Server Example: Multiple Classes


Suppose that an investigation of the database management system log reveals that individual transactions submitted to the database server have significantly different characteristics. However, also suppose that the analyst notes that these transactions can be grouped into three distinct groups of fairly similar transactions, as indicated in Table 2.1: trivial, medium, and complex. These transaction groups differ in the average CPU time and average number of I/Os per transaction. Therefore, it would not be appropriate to characterize the transactions submitted to the database server as a single group. If they were combined into a single group the resulting model may be too approximate and with large errors. Thus, when describing a QN model, one has to also specify the classes of customers that use the resources of the QN, the workload intensity of each class, and the service demands at each resource per class.

Table 2.1. Summary Statistics for the Database Server

Transaction Group

Percentage of Total

Average CPU Time (sec)

Avg. Number of I/Os

Trivial

45%

0.04

5.5

Medium

25%

0.18

28.9

Complex

30%

1.20

85.0

A multiclass QN model should be used in the following cases:

  • Heterogeneous service demands. The requests that form the workload can be clustered into groups that exhibit significantly different service demands on the system resources as is the case in Table 2.1.

  • Different types of workloads. The types of requests in the workload are different in nature. For instance, a database server may be used for both: a) short online transactions and b) a batch workload to generate managerial reports.

  • Different service level objectives. In this case, different classes of requests have different service level objectives. For example, the transaction groups of Table 2.1 may have 1.2 seconds, 2.5 seconds, and 8 seconds, as their acceptable limit on the average response time, respectively.



Performance by Design. Computer Capacity Planning by Example
Performance by Design: Computer Capacity Planning By Example
ISBN: 0130906735
EAN: 2147483647
Year: 2003
Pages: 166

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net