Browse Prior Art Database

Improve LOG DEFER Suspends across CICS WAITs

IP.com Disclosure Number: IPCOM000145200D
Original Publication Date: 2007-Jan-10
Included in the Prior Art Database: 2007-Jan-10
Document File: 2 page(s) / 44K

Publishing Venue

IBM

Abstract

CICS* Dispatcher logic has been enhanced and is now sensitive to the specific type of suspend that is only used for Log Defer suspends. If CICS is about to return control to the operating system because there is no more work to process at this moment in time, AND the next task to be resumed is one suspended for a Log Defer suspend period, then the new logic is executed. CICS already contains and maintains data on the previous few Partition Exit waits back to the operating system. If a previous Partition Exit exceeded 100ms in duration, then it avoids CICS entering the new Partition Exit in such circumstances. The problem this paper addresses is the case when no external event is likely to cause CICS to regain control, and yet there is further work that can be processed soon after. This can result in poor throughput of work because of the excessive waits caused by returning control back to the operating system.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

Improve LOG DEFER Suspends across CICS WAITs

    In an online transaction processing (OLTP) environment, control is returned to the operating system when there is no further work to do, with the expectation that an external event will cause the OLTP system to regain control, and allow further processing. The problem thi paper addresses is the case when no external event is likely to cause the OLTP system to regain control, and yet there is further work that can be processed soon after. This can result in poor throughput of work because of the excessive waits caused by returning control back to the operating system. An example of this behaviour is with an OLTP system such as CICS*, where an optimization is made to get as much data into a log buffer as possible. CICS provides the ability to delay calls to write its log data to the operating system, to allow other tasks within the CICS system to add their own log data into the same buffer. This delay is called the Log Defer suspend interval (LGDFINT). It is possible for CICS to return to the operating system (known as a Partition Exit wait) while suspended for a Log Defer suspend interval. This means control is returned to the operating system for 0.25 seconds (unless an external event POSTs CICS to resume it). Since Log Defer suspends should be of the order of 5ms or less (typically) a 250ms suspend across a Partition Exit is very noticeable in certain customer environments (e.g. where few long-running "batch" style programs are performing recoverable changes overnight, and there is nothing to pre-empt CICS from the Partition Exit WAIT once it has entered it). The approach disclosed here is to use statistical data just prior to the current period to anticipate whether giving control to the operating system is unlikely to be posted by an external event during a defined period of time, and yet further work could be processed soon after giving control ba...