2.7 The Database Server Example: Types of Resources

Suppose that the database server of the previous sections is used to support a client/server application. Client workstations are connected to the database server through a local area network (LAN). Clients work independently and alternate between "thinking" (i.e., composing requests to be submitted to the database server) and "waiting" for a reply from the server. When a reply returns to a client workstation, another thinking/waiting cycle starts immediately. Therefore, we can represent the time spent at the client (i.e., the think time) as a resource that has no waiting line. This type of resource is called a delay resource, which we represent by a circle (without the rectangle that represents the queue).

The LAN that connects clients to servers is an Ethernet LAN, whose effective bandwidth decreases as the number of client workstations increases due to increased packet collisions. Thus, the LAN can be modeled as a resource that has a service rate that depends on its load. This type of resource is called a load-dependent resource and is represented graphically as a circle with an arrow plus a rectangle to represent the queue. Resources, such as the CPU and disk, that have a queue but have a constant service rate, are called load-independent resources. Figure 2.5 depicts the complete QN model with the clients represented as a delay resource, the LAN as a load dependent resource, and database server consisting of two load independent resources.

Figure 2.5. QN for database server with clients and LAN.

To summarize, three types of resources can be used in QN models:

• Load independent (LI). These resources have a constant service rate that does not depend on the load (i.e., on the number of requests in the queue).

• Load-dependent (LD). The service rate of this type of resource is a function of the number of requests in the queue. This type of resource can be used, for example, to model a queue with m resources as in Fig. 2.1 (b). In this case, the service rate increases as the number of requests grows from 1 to m. Alternatively, as in the LAN example above, the service rate of a load dependent server may also decrease as the number of requests grows.

• Delay (D). There is no waiting line in this case. A request that arrives at a delay resource is served immediately. This type of resource is used to model dedicated resources (i.e., resources not shared with other requests) or situations in which there is an ample number of resources compared to the number of requests in the system.