Analyzing MQ Log Manager and Buffer Manager Metrics

Examples of how MQ Statistics data provided by the Log Manager and Buffer Manager components can be used to identify opportunities for improved performance and efficiency.

More MQ Statistics Videos

1. Overview of MQ Statistics and Health Assessments
2. Assessing MQ Buffer Manager Health
3. Assessing MQ Log Manager Health
4. MQ Message Manager Metrics Supplemented by Accounting Data
5. Log Manager and Buffer Manager Metrics
6. Sample MQ Statistics and Accounting Dashboards

Video Transcript

So as is the case with Db2, well-performing MQ logging infrastructure is essential to support recovery and backout without impacting ongoing performance. Logging activities driven primarily by puts and gets for persistent messages, and those represent a minority of total messages in most environments. Nevertheless, the metrics on the volume and drivers of persistent messages are captured in the MQ accounting data as we’ll see later. And so they can be identified from there. So as we touched on earlier, log manager metrics can help identify bottlenecks that might be occurring in log processing. And so again, from a profiling perspective might be helpful to look at the volume of data being logged, as we see here. And we saw earlier in the health assessments that waits for log buffer occurring in the early morning hours and how that correlated with a high volume of log data being written. So let’s examine that more closely.

Again, let’s look at a time-of-day profile on this volume of log data. We do that, and we see that that early morning spike is a very common occurrence. The green bars indicating here, you know, the bottom of that’s around 25 megabytes a second. So almost every day is that amount or higher during that time. So again, let’s go ahead and capture this in the dashboard. All right, let’s go ahead and look again at the volume of log data being written here. And in this case, let’s compare it to the number of log write I/Os. So there’s a pretty consistent relationship during the day between the two and doing a little math on the side, it’s about 5k of data being written per IO. But in these early morning times, we can see that there’s a much higher amount of data being written per IO. In fact, do the math it’s about 68K. So again, that’d be interesting. Fortunately for this site, I don’t have accounting data that would be helpful to research that further. Let’s go and capture that in our dashboard.

All right. So we looked at message manager, log manager data. Let’s go ahead and look at buffer manager data. Earlier in the health assessments, we identified exceptions for Disk read operations and reaching the deferred write threshold around 8:00 PM on that selected day. So again, let’s look at the profile over the month for this particular deferred write threshold metric. And we can see that actually, the early morning hours are a bit more, there are more occurrences on a typical day than we saw during that one day when we saw the 8:00 PM one and the 45,000 we saw was definitely an exception during that time. Typically it’s down at, you know, 3000 or so. Talk about read operations from disk. Again, the negative outcome of flushing the data out of buffers you have when you have to go back and read it from disk. And so again, let’s go ahead and look at a profile of this as well. And when we do that here, we can see that having to read messages from disk occurs far more frequently in those early morning hours. And in fact, on most days, it’s in the thousands per second. Again, that early morning hour tends to be the time in this environment where there’s more stress on the buffer situation.