2.6 The Database Server Example: a Mixed Model

Consider again the database server of the previous sections. Assume now that management has decided that the online transactions of Table 2.1 and the batch management reports should be allowed to execute at the same time at any time of day. However, before management commits to this decision, it needs to know if the database server will be able to support both these workloads while meeting the performance objectives for all of them, as specified in Table 2.2. The performance goal for online transactions are specified in terms of an upper bound on the average response time. No limit on response time is set for the batch workload. However, a lower bound on throughput is specified for that class.

Table 2.2. Service Level Agreements (SLAs) for the Database Server

Transaction Group	Maximum Average Response Time (sec)	Minimum Throughput
Trivial	1.2	-
Medium	2.5	-
Complex	8.0	-
Batch Reports	-	20 per hour

Such performance goals are also called Service Level Agreements (SLAs) since they represent an agreement between service providers and consumers of a service. SLAs and QoS objectives are related and sometimes used interchangeably. SLAs are specified for each class and may involve many different performance metrics such as response time, throughput, and availability. For example, one may have the following sets of SLAs for the services provided by a Web site:

• 99.99% availability during the 8:00 a.m.-11:00 p.m. interval, and 99.9% availability at other times.

• Maximum of four seconds of page download time for all requests submitted over non-secure connections and maximum of six seconds of page download time for requests submitted through secure connections.

• Minimum throughput of 2,000 page downloads per second.

The QN model required to answer the performance questions at the beginning of this section is a multiclass mixed QN model. It is mixed because some classes are open and some are closed. The representation of this model for the database server example is shown in Fig. 2.4.

Figure 2.4. Mixed queuing network for a database server.