GB1452865A - Data processing apparatus - Google Patents

Abstract

1452865 Data processing systems INTERNATIONAL BUSINESS MACHINES CORP 9 Oct 1974 [21 Nov 1973] 43665/74 Heading G4A Interactive terminal jobs and batch jobs are swapped between main and auxiliary storage in dependence on measured and estimated rates of usage of system resources by the jobs and in dependence on performance objectives specified for the jobs. When a new user logs on or a new batch job is selected, or, alternatively at regular intervals, a work load manager is invoked to process two queues of user control blocks so as to enable the performance of the system to be optimized. The user control blocks corresponding to jobs which are in main storage form an IN queue, and the first function of the work load manager is to access, 31, Fig. 2, the user control block at the head of the IN queue via a pointer 17, Fig. 1. From the accessed user CB a performance group number which has been assigned to that particular user is used to index an installation performance specification table IPS. The data stored in this table for each performance group includes constants A-C used in calculating the accumulated number of service units for each job and, for each of a number of periods which may be related to the amount of time elapsed or the number of service units accumulated since processing of the job began, a performance objective in the form of a table of service rates and corresponding normalized load levels. The number of service units accumulated by a job may be equal to A times CPU time + B times I/O count + C times a measure of main storage occupancy, the CPU and I/O count being stored in an address space control block corresponding to that user, and the main storage occupancy measure being a function of page count and CPU time. The service rate is calculated as the quotient of the number of service units accumulated divided by a time interval which is equal to the difference between the CPU time in the AS CB and the old CPU time in the user CB. From this calculated service rate the corresponding normalized load level is determined from the table in IPS and the same procedure is carried out for the next user CB in the queue. Finally the user CBs are reordered according to the new calculated normalized load levels. A similar reordering of the user CBs in the OUT queue is performed except that the queue is in reverse order of normalized level and anticipated service data are used since the jobs are not actually in course of execution. Should the normalized level at the head of the OUT queue be greater than the normalized level at the head of the IN queue then a swap between the heads of the two queues takes place. By providing different performance objectives different users can be handled in different ways and the different service rates specified in each performance objective permit the same user to be treated differently depending on the system work load.