Job Schedulers

We asked eight vendors to send us their job-scheduling software for testing in our Syracuse University Real-World Labs®. Argent, BMC Software, Computer Associates International, Cybermation, Tidal Software and Vexus Consulting accepted our challenge. Hewlett-Packard said it doesn't have a product fitting our criteria, and IBM declined to submit its Tivoli software.

Five of the products we tested--the exception is Vexus Avatar--work similarly. A central scheduling server interacts with a database to store and schedule jobs. When it's time for a job to run, the scheduling server contacts a lightweight agent program, which signals the endpoints that will perform the job. The agent then executes a script and can return status codes and error information to the scheduling server. Additional jobs can be launched, a series of jobs can abort, or the scheduler can wait for an operator to take over. These products support failover to a backup scheduling server. Avatar works a bit differently, with each endpoint using a small, local scheduling server. As such, jobs can run independently without communicating to a central server at all.

Computer Associates and BMC make the most advanced products in this field. Although the difference in their Report Card scores is minuscule, CA Unicenter AutoSys Job Management 4.5 received our Editor's Choice award because it has slightly better access control and a simpler management interface, and it supports a few more client platforms.

Management, job control and reporting capabilities accounted for 95 percent of each product's Report Card grade. We saved only 5 percent of the score for price: We feel the other factors are more important, and the vendors' differing business models--based on such factors as usage time, number of processors and operating systems--make a fair pricing comparison difficult. If your large organization relies on job scheduling as a critical business process, high availability and scalability matter more than price. Conversely, in small environments with only a handful of servers, factors like scalability and role-based administration may not matter at all.Our management category covers role-based administration, job-priority scheduling, management user interface and agent platform support. Role-based administration is especially important for installing a large job-scheduling product, creating groups and users, and granting access. Tidal Enterprise Scheduler, Argent Job Scheduler and CA Unicenter AutoSys can pull users and groups from the corporate directory.

We put heavy emphasis on the main tasks of setting up schedules and prioritizing jobs. Scheduling tasks include creating and combining multiple calendars, dealing with holidays, and deciding when jobs should run. Prioritizing refers to controlling how many resources jobs use. Jobs that are time-sensitive or critical should get higher priority.

All the products we tested let you configure permit and deny calendars. A job runs whenever the permit calendars dictate--every Monday night, once per quarter or every business day, for instance. Deny calendars keep jobs from running and overrule the permit calendars if a job appears on both.

We were most impressed with Tidal Enterprise Scheduler's and Vexus Avatar's management interfaces, which located existing jobs and set job parameters easily.Agent platform support is diverse. Every vendor supports Windows NT and above, and all but Avatar support Hewlett-Packard HP-UX, IBM AIX, Linux and Sun Solaris. Smaller and niche systems, like OpenVMS, Compaq Tru64 Unix and Dequent Dynix, get some support. Only CA's Unicenter AutoSys supports Mac OS X Server, a Unix derivative. Schedulers from CA, BMC, Cybermation and Tidal all offer mainframe support.

Job Control

In testing these programs, we focused on job-control tasks, such as prerequisite checks, job creation and error recovery. Job schedulers don't actually create the batch files to run on the end nodes; that's up to the IT staff. Instead, schedulers kick off a script, batch file or executable at the designated time. These programs must check for prerequisite conditions before running the job--say, checking disk space before starting a backup--and handling error recovery if the job fails. We were disappointed with the prereq capabilities of our top players, CA Unicenter AutoSys and BMC Control-M. They were limited to file checks and checking the status of a previously run job. By comparison, Cybermation's ESP Espresso had the best procedures for checking prerequisites. It could detect file presence and size changes; monitor the event log, text strings, processes, services, CPU utilization and disk usage; and perform SQL queries.

Creating a single job is relatively simple. Just give it a name and specify which command to run on which server. For job creation, we graded each package's ability to create and visualize complex environments with multiple jobs running on various servers. CA's Job Visualization add-on lets you see all jobs programmed into Unicenter AutoSys. When you click on any job, you can see all possible paths to that job and all paths after it. Without Visualization, you can't graphically see relationships between jobs at all.

In the error-recovery subcategory, BMC Control-M excels. If a job fails, Control-M provides a range of options, such as rerunning the job, changing variables, sending out alerts and running other jobs. You can set multiple branching options for error conditions and make extensive use of command error codes. Argent Job Scheduler likewise has good error recovery: You can act upon a range of error codes, rather than individual codes. Argent's product also tries to rerun a job a specified number of times over a set number of minutes or hours.Cron + Perl != Job Scheduling

Why spend a quarter-million dollars on these programs when you can use cron, Perl scripts, SSH and a little shell programming?

If you just need to run basic standalone jobs, cron might be sufficient. However, cron has limitations that a dedicated job-scheduling product can overcome. Cron is like an alarm clock. At a certain time, it wakes up, kicks off a job and goes back to sleep until the next job comes along. You can't correlate one job to the next. Cron can't tell that your log-file rotation failed at 2 a.m. and you shouldn't delete the old log files at 4. It can't tell if a job has finished early or late, or whether to move the rest of the jobs up or down. Homemade solutions, meanwhile, may involve setting up checkpoints and writing additional script files, and ultimately will pose scalability and longevity problems. It's also hard to see the status of cron jobs when you're dealing with many interdependent servers.

The job schedulers we tested are a bridge between mainframe and Unix environments. You can now get mainframe job-management functionality on your Unix, Linux or Windows server. Vendors that have mainframe backgrounds or support mainframes in their job-scheduling suite fared better than the rest.

CA entered the job-scheduling field almost three decades ago, so it's no surprise that this version of Unicenter AutoSys Job Management reflects years of enhancements and experience. The package has an easy-to-use graphical interface for Unix and Windows admins, as well as a Web client for operators and technicians. The interface is one reason Unicenter AutoSys edged out BMC's functionally similar Control-M.Management settings are found in the administration program, a Web-based operator's site, and, optionally, eTrust Access Control. We used the Web interface to create simple jobs, kick off new jobs and check the status of scheduled events. The eTrust software provides granular read, write and execute permissions on all aspects of the job-management suite. We could control jobs, calendars, machine access and reports on a user or group basis. Credentials are provided by native Windows authentication. AutoSys includes eTrust, but configuring that program isn't simple. If you don't install it, though, you can't take advantage of group access controls under Windows.

We easily managed and scheduled related jobs from the main administration interface. To create a job, we specified a job name, the job owner, the command, dependencies and the machine. We liked the product's use of boxes for grouping jobs--a kind of batch job of batch jobs. Kicking off a box runs all jobs inside that box simultaneously, unless a job depends on the completion of a previous job.

Unicenter AutoSys has a unique method of failing over to a secondary job scheduling server: It uses a third machine for control purposes. This third server requires almost no resources--just a simple agent program that receives heartbeats from the job-scheduling servers. If you have the primary and backup scheduling server in different locations, the third machine determines if the primary actually failed, or if the secondary machine's network connection died. The three machines send one another heartbeat messages to confirm that the remote systems are up and the job scheduler is running. If the secondary machine can't reach the primary but can reach the third machine, it takes over as master. However, if the secondary can't reach the primary or third, it assumes that the problem is on the secondary's end and doesn't take over. Switching back from the secondary to the primary requires manual intervention.

You can configure Unicenter AutoSys not to use a third machine. In environments where job scheduling is critical or the backup server is separately located, the third machine is a powerful tool.

Unicenter AutoSys Job Management 4.5. Computer Associates International, (888) 423-1000, (631) 342-6000. www.ca.comBMC and CA job schedulers are nearly equal in function and earned similar scores in our Report Card. BMC's Control-M for Distributed Systems offers excellent calendaring capabilities, late-job predictions and error recovery. However, the product is harder to administer than Unicenter AutoSys. Control-M has a Web interface for running and viewing jobs that helped raise its management score, but we prefer AutoSys' interface.

As with CA's product, we could use Control-M to bundle multiple jobs into a larger group. Creating a dependency between two jobs, such as running a log-rotate script before running a log-backup program, is as simple as dragging one job on top of another. Unfortunately, we couldn't specify file-existence, disk-usage, process-running or any other system-level dependencies. These functions must be handled within the batch scripts.

Predefined templates, which BMC calls skeletons, simplify job creation--individually or en masse--by propagating configuration fields automatically. Updates to the templates are sent to all jobs that use it.

Control-M provides multiple exit conditions for failures. The job can be rerun, a global variable modified, alerts thrown, e-mails sent or other jobs started. We specified the exit code and used text strings to define failures, and could detect and act when the Unix file copy command returned "No such file." Control-M integrates with BMC Patrol for alert management and automatic failover, though you don't need the Patrol suite for basic alerting. Unfortunately, without Patrol, failover from one scheduling server to a backup is a manual process. Built into Control-M is an alert console, similar to CA's scheduler. Alerts, which may be e-mailed, can be indicated as acknowledged and handled.

Control-M uses time heuristics to determine if a stream of jobs will be late. The software keeps tabs on how long each job is supposed to run. If a job upstream is running behind schedule, threatening to make a downstream job late past a certain threshold, the application triggers an alert. An operator can abort the job stream, halt other less vital jobs or find out why things are running late. The other products we tested determine lateness on a job-by-job basis.Control-M. BMC Software, (800) 841-2031, (713) 918-8800. www.bmc.com

The excellent management interface on Enterprise Scheduler made this package one of our favorites to use. Tidal's suite also has the best documentation, with real examples and step-by-step instructions, but we want better reporting, job creation, visualization and error recovery.

To assign access control, Enterprise Scheduler uses security policies, which contain all the permissions you want to set. You assign users or groups to the policy from Active Directory. With this setup, it was easy to modify policies later and have the changes affect all relevant users.

As expected, Tidal's scheduler can create dependencies based on previous job status and global variables. The variables may be a string, number, date or Boolean value. A job can modify or read the variables. You also can create file dependencies, such as a log-rotation job that acts only when a log file larger than a certain size is present. Within a list of dependencies, we required all to evaluate as true, or at least one. However, we couldn't specify for at least two dependencies to be true, or create complex "and/or" statements.

The system can detect errors in three ways. Enterprise Scheduler supports exit codes, even letting you specify a range of exit codes for success or failure. A program's output can be piped to another program, and the error code of the piped program is used to determine success. Finally, you can do a pattern match on a program's output for success or failure. Although these functions can be handled inside the batch job, having it available inside the job scheduler is a bonus.Tidal Enterprise Scheduler 5.0. Tidal Software, (877) 558-4325, 650-475-4600. www.tidalsoftware.com

A mixed bag, Cybermation ESP Espresso does the essential tasks of creating jobs and checking dependencies well, but this package needs better alerting and scheduling features. Its interface is great for some tasks and poor for others. It's easy to create jobs and visualize the job stream, for example, but the general system interface is difficult to use.

The job-creation component is put together well. Icons representing jobs can be dragged and dropped onto a canvas like a Visio diagram. This lets you see dependencies and the order of operations for a job; you can also monitor jobs in real time from this view.

ESP has the best dependency checking of the products we tested. We could monitor the event log, text strings in a file, processes running, services, CPU utilization and disk usage. We also could perform SQL queries. You can launch jobs based on the results of these monitors.

Access-control settings are available on a per-user or group basis. We copied permissions from one group to another and from one user to another. Permissions can be granted to jobs and calendars as well. For example, we let a user modify all calendars except for the "payroll" one.ESP Espresso 4.2. Cybermation, (905) 707-4400. www.cybermation.com

With strong reporting and alerting engines, the Windows-only Argent Job Scheduler may be a good fit for small and midsize environments. To appeal to large enterprises, though, it needs better support for dependency checking, role-based administration and job-stream visualization.

The software pulls user and group information from the Active Directory domain or local computer when no domain is present. You can grant read, write, execute and control permissions per machine or per job class. A job class is a grouping of individual jobs; you can't set controls on individual jobs or calendars. This setup makes it difficult to visualize job dependencies, even though job classes can be used to organize job streams.

Argent's reporting and alerting, sent over e-mail or pagers, are better than the competition's. The alerts can play a sound for a set length of time, execute a command on the scheduling server's command line or send a Windows message alert to any Windows client. Alerts also can be sent to Argent Guardian, a separate monitoring and alerting product. You don't need Guardian for basic alerting. Reports can be created and sent on a calendar schedule as well.

The Argent Job Scheduler 4.5A. Argent Software, (860) 674-1700. www.argent.comVexus Avatar has two benefits: It's simple and inexpensive. Avatar doesn't offer many configuration options, and creating jobs with it is straightforward. The price runs as low as $500 per CPU. Vexus claims Avatar can compete at the same level as Unicenter AutoSys or Control-M, but we found the product's distributed setup a disadvantage in large environments. Avatar is best-suited for those looking for small-scale or simple batch management, or on machines that have limited communications with a central job scheduler.

Avatar is vastly different from the other products we tested. Instead of working with a model of dumb agents with a central scheduling server, Avatar agents operate independently. A full, yet lightweight, scheduling server is installed on every endpoint, which holds its own calendar, queues and job definitions. Agents are administered through a separate Web client, which connects to an Avatar server for administration only, not for triggering tasks. Jobs can be triggered across machines, and you can add multiple Avatar servers to a management interface, but there's no central machine for the actual job creation.

Avatar doesn't provide a mechanism for shared security settings, calendar, groups or jobs. Users authenticate against the local system password file or directory. It does not support groups, so access controls are set for individual users or for every authenticated user. Each machine can set "allow" or "deny" rules for file transfers, executing and monitoring jobs and setting dependencies on a per-host or -user basis. Jobs can be triggered across machines, but Avatar lacks intramachine visualization tools.

Avatar Job Scheduling Suite 4.5.5. Vexus Consulting Group, (416) 985-8554. www.vexus.ca

Michael J. DeMaria is an associate technology editor based at Network Computing's Syracuse University's Real-World Labs®. Write to him at [email protected].Keeping your systems running on time requires coordination of tasks, from backing up servers to generating sales reports. Although you can concoct an application to start operations and keep tabs on each job, the six job-scheduling suites we tested at Network Computing's Syracuse University Real-World Labs® automate the process.

We tested products from Argent, BMC Software, Computer Associates International, Cybermation, Tidal Software and Vexus Consulting, and graded their job-control and -reporting capabilities, as well as the ease with which we could manage the programs.

All the suites but Vexus' Avatar require a central server and database, in which the server uses an agent to trigger the start of each job on an endpoint. Avatar stores a scheduler on each endpoint and is better suited to small-scale operations.

The CA and BMC products earned very close Report Card scores for their sophisticated scheduling abilities. But CA's Unicenter AutoSys Job Management 4.5 won out for Editor's Choice because its management interface and extensive platform support were slightly better than those of BMC's Control-M.

To test the job schedulers, we used a dual 2.4-GHz Pentium Xeon system with 1 GB of RAM, running Windows 2000 Server SP4 as our job-scheduling server. A second machine ran as a backup. A 600-MHz Pentium III system with 256 MB of RAM running Windows 2000 Server SP4 served as our clients. If the vendor didn't supply a built-in database, we installed Microsoft SQL 2000 Service Pack 3a. For Linux tests, we used a Red Hat 9 system and installed some components of Vexus Avatar.Our scheduled jobs comprised batch files and command-line executables, and we added a sleep command to some of the batch files to make some jobs run late. To determine how well a product can handle failure conditions, we made batch files terminate with nonzero exit codes.

All Network Computing product reviews are conducted by current or former IT professionals in our Real-World Labs® or partner labs, according to our own test criteria. Vendor involvement is limited to assistance in configuration and troubleshooting. Network Computing schedules reviews based solely on our editorial judgment of reader needs, and we conduct tests and publish results without vendor influence.

R E V I E W

Desktop Management

Sorry,
your browser
is not Java
enabled

Welcome to NETWORK COMPUTING's Interactive Report Card, v2. To launch it, click on the Interactive Report Card ® icon above. The program components take a few moments to load.

Once launched, enter your own product feature weights and click the Recalc button. The Interactive Report Card ® will re-sort (and re-grade!) the products based on the new category weights you entered.

Click here for more information about our Interactive Report Card ®.