Many years of Managing performance on the System i (As/400, iSeries, i5 e.t.c.) has meant having to get used to data overload. Green screen textual presentation of the relevant metrics. Yes of course We’ve all used Excel and downloaded the data and made some pretty and pertinent graphs, however, the crunching of these stats meant coding quite a few lines to output the data, as you may want it.
Further more, the reports in performance tools are all really good and the output from PEX again is very good, yet very demanding to drill down to a problematic transaction. It takes a lot of time to get to the point, by the time you have, then there’s no point, if you see what I mean.
So, a visit to the ‘i’ Doctor is necessary. The ‘i’ Doctor I mean is IBM’s iDoctor offering, comprising of Job Watcher, PEX analyser and Heap analyser. In this article I will only concentrate on the Job Watcher component.
Job watcher is a god send for any System i administrator responsible for performance and capacity planning. Why? Well it’s your job to maintain performance and to assure adequate resource by capacity planning and Job Watcher is an ideal tool to assist.
By taking relevant samples, for instance, during your peak periods and using these as baselines, you can simply identify deviances beyond your baseline and obviously keep them in check. That’s just one use. The other use really is when you have that elusive performance problem, then that’s when Job Watcher really does come into it’s own.
At the first sign of a performance issue, just submit a Job Watcher sample, which will record all the waits for all the jobs within your sample range. You can select all jobs running or filter the sampling by many different criteria.
Once the sample is taken, you summarise the data by taking an option, then as if by magic you have an amazingly detailed graphical view of the waits, which make up all the transactions running on your System i. What more could you want? Well, there is more, lots more. One exciting feature that has recently been added is a pie chart overview of the sample period showing how all the waits on the system are made up.
All transactions have some element of waiting, CPU Queuing, Seizes, File Locks, Record Locks, DASD Paging, DASD writes, DASD reads and so on. Some of the waits you see may be ‘good’ waits, some may well be ‘bad waits’ Job Watcher show them all.
Once you display the Wait Graph by interval, you will surely see the culprit spike in the graph. From there you can drill down to display the actual jobs involved in the wait. The actual waits per job, if it’s a record lock you can see the holder and the waiter and even drill down to the program stack of each waiter and holder. Surely that is enough information?
At this point you now have the job’s program that is waiting, the program that is holding and even the record number in the file that is being waited upon, wow.
Without going in to a full tutorial here, it is impossible for me to cover Job Watcher’s amazing value in time saving, capacity planning, performance enhancements and the offshoot revenue enhancements delivered from removing the bottlenecks. For more information and a tutorial visit TheiSpecialist.com
See you in the waiting room at the ‘i’ Doctor.