Real-time computing. Real-time news.
November 12, 2007
A Job Scheduler for the Ages
1 | 2 All »
More than a Job Scheduler
It’s funny how we argue over semantics in the world of distributed computing. Whether a particular software product creates a grid or a fabric or whatever varies depending on whom you ask and, quite frankly, under which buzzword its vendor chooses to market it. Therefore, one might be a little underwhelmed by Advanced Systems Concepts Inc.’s (ASCI) decision to market its ActiveBatch software as a “job scheduler,” even though its list of capabilities lends the product to far sexier (and more nebulous) labels.
ActiveBatch, branded from its inception in 1997 until 2000 as Batch Queue Management System, is geared toward job automation over distributed and heterogeneous platforms, including those running Windows, Linux, IBM AIX, HP-UX, Solaris and/or OpenVMS operating systems. The only requirement in terms of operating systems is that the job scheduler must be on a Windows machine. As for its distributed nature, ActiveBatch has tested to be scalable up to 2,000 server connections, and up to 1.3 million jobs per day. In terms of what kinds of ActiveBatch can handle, pretty much anything (scripts, applications, file operations, etc.) is fair game so long as it can support a command line -- and no rewriting or recoding is required.
Where ActiveBatch really earns it stripes, though, is in the breadth of scheduling and automation options it offers. According to Jim Manias, vice president of sales and marketing for ASCI, jobs can be scheduled to run simply by time or by a variety of triggers, including the completion of other jobs or the presence of a particular file (or the lack thereof). Going more granular, specific exit codes from completed jobs will trigger corresponding jobs to begin, and when setting file-based automation, jobs can be programmed to be triggered by whether or not a file contains specific attributes.
In addition, ActiveBatch’s workload balancer can send jobs to either to predetermined machines or to available machines, and to either one machine or to a group of machines. It also can handle transaction processing based on user-defined criteria, such as the size of the queue or the time of day. And the product’s Job Steps feature opens the doors to parallel processing by allowing users to decompose jobs into various parts that run on separate machines at the same time. New in Version 6 is a feature called Virtual Root, which allows for multi-tenancy of the job scheduler, enabling transparent distribution of jobs over various departments within an organization. It is event-driven processing at its finest, and all designed with the goal of maximum efficiency or, as Manias puts it, reducing “slack time.”
“Twenty-four hours is not what it used to be,” he said. “People have less time … to be able to deal with ensuring that they achieve the success that they sign up for in terms of service levels.”
Also very important to ASCI is the ability to resolve, or at least mitigate, problems as quickly as possible. Aside from server to server failover, ActiveBatch works with Windows Management Instrumentation (WMI), a collection of about 15,000 hardware and software commands within the Windows OS that allow users to register certain events. Manias gives the example of a user registering the termination of a specific service with WMI, wherein the alert generated by WMI would trigger ActiveBatch to restart the terminated service. One customer, he added, has ActiveBatch programmed to page security to the server room should the intrusion switch on a server be flipped between 7 p.m. and 7 a.m.
Another key aspect of this mitigation strategy is to ensure that jobs get done correctly and on time. To accomplish this, ActiveBatch employs checkpoints that allow jobs to automatically restart from the last checkpoint in case of failure. The solution also incorporates runtime monitoring, which compares the current runtime to historical averages. If, explained Manias, a job that normally takes three hours were to complete in 10 seconds, ActiveBatch would assume something went wrong and rerun the job.
“Everything is done in what I refer to as ‘near real time,’” said Manias. “I would not use it to monitor carbon rods in nuclear reactors, but certainly in the near-real-time world which makes up today’s global economy, it has absolutely been proven to be a top performer.”
A Wide Range of Customers
Page: 1 of 21 | 2 All »
Article Tools
Share Options
(Digg, Technorati, more)
