Gotcha! Deploying Job-Scheduling Software

Companies need software programs, called job schedulers, to control what programs are going to run and when, to keep orders handled and operations humming.

But this doesn't necessarily happen the moment an order is placed.

"There's a misconception that just because the real-time enterprise is here, all processes will go real-time," says Gartner Inc. analyst Milind Govekar. "But batch has been around since about day two in computing, and it's not going away."

To minimize use of its network, a company may send orders in batches at set times to a distribution center.

At Eckerd, updates to its data warehouse—its central data repository—are scheduled to run nightly, after stores have reported their sales for the day. Eckerd uses IBM's Tivoli Workload Scheduler, but there are other job schedulers, such as BMC's Control-M, that also can ensure each job is handled successfully.

For example, existing job schedules required too much "hand-holding," according to Eckerd officials, due to "dependencies" between jobs. Dependencies exist wherever one job must be completed before the next job can begin. Such dependencies tend to accumulate as new jobs are programmed into a scheduler over time and it can be easy to lose track of them.

At Eckerd, sales data sometimes wasn't sent to the data warehouse for two or three days because the process of gathering and consolidating sales data frequently stalled and had to be restarted by data-center computer operators. Job scheduling software works by launching multiple programs or batch scripts in the correct order to achieve an overall result—in this case, an update of the data warehouse. If the program for processing pharmacy insurance claims fails, for example, pharmacy sales totals won't be available to feed into the overall sales results.

Eckerd's software team programmed the scheduler to log when a job took longer than expected. This allowed Eckerd to analyze the dependencies that were causing the process to break down and concentrate on solving those problems. Today, Eckerd's data warehouse is updated every night.

If multiple administrators are doing their own scheduling at different sites or on different installations of the software, tracking dependencies is even harder.

Steve Farrell, an operations manager in Merrill Lynch's mainframe data centers, recalls that, a few years ago, the broker had three mainframes with their processing controlled by separate organizations. A dependency breakdown "did stop one data center in its tracks waiting on a feed from another."

The incident taught Merrill's data center managers they needed to do a better job of documenting the relationships between jobs running at different sites, says Farrell.