Managing Interactive User Applications and Batch Jobs in a Cluster Environment

In the previous discussion, you saw two types of users: interactive users who perform many small I/O operations from the keyboard, and users who run batch jobs and database reports.^[18] Each type of user places different demands on the cluster node CPUs and on the NAS server. The first type wants to use the CPU and the NAS server briefly and intermittently during normal business hours, while the second wants to grab as many CPU cycles and NAS I/O operations as possible for long periods of time. To satisfy both types, the cluster administrator can isolate each type of user application on two different cluster nodes.

Note 

Historically, on a monolithic server, the Unix system administrator reduced the impact of batch jobs (CPU-and I/O-hungry programs) by running them at night or by starting them with the nice command. The nice command allows the system administrator to lower the priority of batch programs and database reports relative to other processes running on the system by reducing the amount of the CPU time slice they could grab. On a Linux Enterprise Cluster, the system administrator can more effectively accomplish this by allocating a portion of the cluster nodes to users running reports and allocating another portion of the cluster to interactive users (because this adds to the complexity of the cluster, it should be avoided if possible).

Run Batch Jobs Outside the Cluster

The NAS server, however, is still a single point of contention for these two types of applications, so you'll want to purchase the fastest NAS server you can afford. Another way to reduce the contention for the NAS server, if your budget doesn't allow you to purchase one that is fast enough to handle both types of users at the same time, is to use the rsync utility to make a nightly copy of the data stored on the NAS server and place this copy of the data onto local storage on an old server. You can then run batch jobs that do not require up-to-the-minute data on this server (month-end accounting reports can use the rsync snapshot of yesterday's data, for example).

Use Multiple NAS Servers

Of course, you can also eliminate the NAS server as a single point of contention by using more than one NAS server and spreading your data across two or more NAS servers. For example, the central data center for a multi-warehouse facility could store the data for each warehouse on a separate NAS server.

However, if your budget allows, the easiest configuration for the system administrator to build and maintain is the one that lets users store all of their data in one place: on a single NAS server. And how fast does the NAS server need to be? In the next two sections, we will look at a few of the methods you can use to evaluate a NAS server when it will be used to support the cluster nodes running legacy applications.

^[18]Possibly batch jobs that run outside of the normal business hours.