2.8 The Database Server Example: Blocking

Suppose now that the IT managers of the database service want to provide a response time guarantee to their customers. In order to provide this guarantee, regardless of the arrival rate of requests, the number of concurrent database transactions has to be limited. In other words, some form of admission control has to be implemented (see Fig. 1.4). With an admission control, a limit, W, is set on the number of transactions allowed into the system. An arriving transaction that finds W transactions in the system is blocked (i.e., it is either rejected or placed in a queue waiting to enter the system). This is illustrated in Fig. 2.6, which shows the variation of the number of requests in the system versus time. Each arrow pointing upwards indicates an arrival. At arrival instants, when the load is less than W, the number in the system increases by one. Downward arrows indicate departures; at these instants, the number in the system decreases by one. The picture shows five requests arriving when the number of requests in the system is at its maximum value, W. These requests are blocked. Examples of mechanisms that limit the number of requests in a system include the maximum number of TCP connections handled by a Web server, the maximum number of database connections supported by a database server, and limits on multiprogramming level enforced by the operating system.

Figure 2.6. Number of transactions in the database server vs. time with admission control.

Limiting the total number of requests allowed in a computer system limits the amount of time spent by requests at the various queues and therefore guarantees that the response time will not grow unbounded. This is a type of congestion or admission control. Clearly, if requests are rejected, they will have to be resubmitted at a latter time. Figure 2.7 shows the QN for the database server with admission control indicating that transactions are rejected when the system is at its limit. It should be noted that in this case, the throughput is not necessarily equal to the arrival rate. In fact,

Figure 2.7. Database server QN with admission control.