Db2 Accounting: Exploring Prefetch Activity and Suspension Events

Examples of the benefits of dynamic navigation through Db2 Accounting data applied to metrics such as (a) prefetch activity by type and (2) counts of suspension (“wait”) events.

More Db2 Accounting Videos

1. Exploring Analysis by Connection Type
2. Exploring Analysis by Correlation ID
3. Exploring Elapsed Time Profiles
4. Exploring Analysis by Authorization ID
5. Exploring Prefetch Activity and Suspension Events
6. Exploring Analysis by Plan or Package Name
7. Exploring Database Sync I/O Activity
8. Case Study: Isolating Change Drivers
9. Exploring Other Metrics in Accounting Data by Plan
10. Accounting Data: Customized Dashboard Recap

Video Transcript

All right, earlier when we viewed the various types of prefetch activity by connection type, we saw some increases in sequential prefetch activity. And so now that we have talked about exploring at the auth ID level, let’s do that for that early morning spike. And we can see that this auth ID is generating most of the sequential prefetch requests, which happens to be the same auth ID that we just saw earlier with those CPU spikes. So again, let’s view this over time and broaden the time interval out to the entire week. And so again, we see this process on a daily basis generating significant volumes of sequential and dynamic prefetch activity and those early morning hours on a daily basis. So, again, this is all kind of correlating together for this activity driven by this auth ID.

Another metric that’s present in the accounting data is the count of suspension events. Db2 performance experts like John Campbell, advise analysts to be aware of both the number of suspensions and the suspension time. And they’re both captured in the accounting records. We looked earlier at the elapsed time, and now we’re going to look at the counts. So, you can see here that Db2 captures approximately 30 types of waits here. And so let’s go ahead and look at this over time. You can see in this environment, the DDF work, the two primary suspension events are waits for database sync I/O and global contention for child L-locks. But notice this pocket of waits for other read I/Os which as we talked about earlier often mean waits for prefetch I/Os. And notice the timeframe here. Does any possible correlation come to mind?
Well, let’s go back and look at the suspension events by auth ID, and lo and behold, we’ve got that same auth ID generating lots of waits for other read I/Os or lots of prefetch activity. And again, if we view these suspensions over the entire week, we see them occurring at the same times as the CPU spikes and the prefetch spikes. Again, one of the things we learn as we explore the data is we can begin to correlate different types of activities and how they relate together. In this case, it’s all being driven by this same auth ID.