Moving Workloads for Server Consolidation

Moving Workloads for Server Consolidation

One of the new options for Patrol for Performance version 6.6.00 and going forward, is the ability to move the total workload from one node to another node. What will save much time is the ease of moving I/O and memory consumption. The procedure is outlined below:

When you are in Predict, you save the workload that you want to move to the library. Once it has been saved or exported to a library, you will need to change the following two lines in the workload file (.wkl):

Example of the section that needs to be modified

To move the Workload, I/O and memory consumption, make the following changes:

SET ADD-FROM-LIBRARY WKLOAD #WKL-NO-DISTRIB

SET NODENAME <nodename>

Note that just the comments were removed from the beginning of the two lines. Do not remove the comment from “#WKL-NO-DISTRIB”.

Also you will need to change <nodename> to the node that you are moving the Workload information to.

Now just import the workloads into the model that you want to move them to.

What will occur is that all the disks that were on the node that you were moving the Workload from, will now be created on the new node. This in turn will move only the I/O and memory consumption that this Workload was using to the new node.

If you would like to move just the Workload and not the I/O and memory consumption, remove all the comments as shown below:

SET ADD-FROM-LIBRARY WKLOAD WKL-NO-DISTRIB

SET NODENAME <nodename>

Note that just the comments were removed from the beginning of the two lines plus from the middle comment from the first line.

You will still need to modify <nodename> to the name of the node that you are moving the Workload to.

If you don't follow this procedure, you will receive many lines of errors when you attempt add (or import) this workload into a model. And the workload will not be added successfully, either.

There is GUI support for this in 7.1.01, so the file modifications will no longer be required then.

This functionality is really useful -- you get easy moving of I/O and memory consumption between models, which before required command files. Now you need to change a couple of lines (once), and everything works pretty smoothly.

There is also a failure with regard to a remaining error message like "rz6 on node alpha1 is unknown disk". This is supposed to be addressed in 7.1.01 also.