Wednesday, May 19, 2010

DASD Considerations for OMEGAMON DB2

So far we've discussed DASD monitoring considerations for OMEGAMON z/OS and OMEGAMON CICS. Now let's consider OMEGAMON DB2.

There are multiple aspects to consider when we are talking about OMEGAMON DB2. OMEGAMON DB2 collects DASD relevant data for such things as DB2 log I/O, EDM pool I/O information, and object I/O data when doing drill down analysis of virtual pool statistics. This information is provided essentially out of the box, and does not have major overhead considerations.

There are some areas where DASD and I/O monitoring can add overhead, and you do have the ability to control if the data collection is on or off. The first major facility is Object Analysis. Object Analysis is an I/O and getpage analysis facility that will look at all the I/O and getpage activity being done on the subsystem, and correlate that getpage and I/O activity by object, and by DB2 thread. Object Analysis does have an ongoing cost of collection. It does not use DB2 traces, but if the Object Analysis collector is allowed to run all the time, it will add to the CPU usage of the OMEGAMON DB2 collector task. In some shops this is not a big issue, in other (usually larger) shops, it is a consideration. You can optionally configure Object Analysis so that it is off by default, but you may start it, as needed. This is a good strategy for those who want to reduce the cost of monitoring, but still have access to the data when needed. I had an earlier post that describes how to configure Object Analysis to achieve this. Another option to consider with Object Analysis is the thread correlation option. If this is enabled, Object Analysis will use more resource, but I find the thread data to be quite useful.

For those DB2 data sharing users, there is another option to consider, Group Object Analysis. If Group Object Analysis is enabled, that means you are running Object Analysis at the level of each DB2 subsystem (i.e. member) within the data sharing group. That means you have the ongoing cost of running Object Analysis at that level, plus you also have the cost at the level of the OMEGAMON DB2 agent task of correlating the data, in particular if thread correlation is enabled. Group Object Analysis is useful data for a data sharing shop, but understand that you will be pulling potentially a fair amount of data on an ongoing basis.

Now let's consider history. In addition to the Accounting and Statistics trace options in the Near Term History setup options, you also have the option to enable such things as SQL, Sort, and Scan data collection. My recommendation, in general, is to set the Scan option to off. The data will give you an indication of some aspects of scan activity done by the thread, but be advised this data is collected via DB2 IFCID traces, and may add more cost to running Near Term History.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.