IBM Systems Magazine, Mainframe - January/February 2014 - (Page 32)

TECH CORNER Improving SYSOUT PERFORMANCE JES2 SAPI POST optimization can yield results R ecent reports of performance problems related to SYSOUT work selection center around SYSOUT taking longer to process due to an outage in the back-end SYSOUT processing. When it is eventually restarted, JES2 spends so much CPU on SYSOUT selection that the system has trouble working through the accumulated SYSOUT backlog. Kevin Kathmann is an advisory software engineer working in JES2 development since 2008. He has worked more than 20 years with IBM. Alexei Pytel is an advisory software engineer working in z/OS JES2 development since 2007. He has been with IBM for 19 years. Steve Simonson is a senior software engineer working in JES2 as a developer since 2007. He has been with IBM for 27 years. JES2 performance statistics show that for many installations, SYSOUT work selection is a significant part of JES2 CPU cost. SYSOUT work selection happens when a process such as a JES2 printer or an application using the SAPI interface requests SYSOUT from JES2. The SYSOUT work selection has two main parts: characteristics and goes to sleep (i.e., waits on an ECB). When new SYSOUT work becomes available, JES2 finds all waiting applications with matching criteria and wakes them up (i.e., posts the ECBs). This is called SYSOUT POST processing. In V2R1, this part of SYSOUT work selection has also been improved. This optimization is described in this article. 1 When an application starts, it A New Problem requests ready-to-be-processed work from JES2. This is called SYSOUT GET processing. In response to a GET request, JES2 scans the SYSOUT queues and finds the SYSOUT that matches criteria specified in the request. Significant effort has been spent in JES2 V2R1 to optimize GET processing for SAPI applications. This topic will be covered in a future article. 2 When all readily available work has been processed, JES2 returns an indication to the application that no more SYSOUT with matching attributes is available. At this point, the application registers its interest in the SYSOUT with certain SYSOUT GET processing could be compared to a database query: Find a record (piece of SYSOUT) that matches a query (SYSOUT selection expression). SYSOUT POST processing solves an opposite problem. Given a record (new piece of SYSOUT) and a set of queries (processes with their selection expressions), find all queries that match this new SYSOUT and wake up associated applications. Another way to describe this problem is a subscription model. A set of applications subscribes to be notified when new SYSOUT becomes available. Each application has its own selection expression and only wants to be notified about pieces of SYSOUT that match its criteria. In prior releases of JES2 the SYSOUT POST processing was straightforward: 1 Look at each application waiting for SYSOUT 2 Compare its selection criteria to a new piece of SYSOUT 3 If there is a match, POST associated ECB 32 // JANUARY/FEBRUARY 2014 The evaluation of a single selection expression is not particularly expensive. However, this logic does not scale well. As systems become larger, more SYSOUT is processed by the system and more applications are waiting for SYSOUT. Therefore, more CPU cycles are spent on testing for a match between this new SYSOUT against all selection expressions. Something had to be done to reduce this overhead. The Solution-Selection Expression Cache Clearly, there was a need to organize selection expressions used by the applications in such a way that the number of

Table of Contents for the Digital Edition of IBM Systems Magazine, Mainframe - January/February 2014

Table of Contents
Editor's Desk: The father of innovation
IBM Perspective: Mainframe innovation continues
Trends: RDz family of tools helps organizations address new paradigms and business expectations
IT Today: Flash Express integrates secure, high-performance storage within System z
Case Study: Weather or Not: Met Office makes a cost-conscious and test-driven decision to consolidate Linux instances on the mainframe
Cover Story: The System z influence: How mainframe innovations have helped shape and reshape the IT world
Feature: Always moving forward: IBM's System z platform has many more tricks up its sleeve
Tech Corner: JES2 SAPI POST optimization can yield results
How To: How to successfully implement zBX hybrid computing
Solutions: Attunity Replicate 2.1, InSync Plugin, Storage Director 4.1
Stop Run: Coin collecting and history fascinate mainframe strategist
Reference Point - Global Events, Education, Resources for Mainframe
2014 Mainframe Solutions Edition Product Index

IBM Systems Magazine, Mainframe - January/February 2014