ProTech’s RMF REDUX is the newest software offering from ProTech and will analyze WLM data to finally provide a view of system behavior that truly reflects the policies implemented.
While many products focus on interval reporting and real-time analysis, they all lack the ability to indicate how well a policy is actually performing. Since there will always be “ups” and “downs” regarding a particular service class, no general policy decisions can be formulated based on a single interval or set.
When goals are missed or exceeded, it is important to see how these events fit into the overall behavior of the policy and whether they reflect issues that need to be addressed. The data is taken from the SMF type 70 and 72 records and is evaluated based on the number of recorded samples to provide a global view of how work is performing within their respective service classes.
One of our clients was looking at WLM reports from RMF REDUX 20 minutes after he began the software installation!
RMF REDUX sample reports are available.
Using the data from this report, we can see how the analysis might proceed in evaluating this service class. At a glance, we can see that 81% of the time, the goal of 50% was exceeded, although with the small number of samples and the high percentage of idle time, we can see that the service class is not very active. Given the small amount of execution time, this service class is not a particularly good candidate for velocity management, since there are so few samples on which to base decisions. This is a good example of a service class where the work might be more suitable for the SYSSTC definition, since it is a small CPU consumer with a high percentage of idle time.
In this example we are looking at an extremely busy service class. With a velocity goal of 30%, it exceeded its definition 70% of the time. It should also be noted that because the goal is only 30% and it is being exceeded most of the time, there was one instance where WLM actually resource capped this service class even though there is no Resource Group defined for it. We can also note that this service class is consuming nearly three (3) engines with an effective utilization of 294.1% within this LPAR. Therefore a significant concern would be whether our goal of 30% is too low to be satisfactory should system resources become more constrained. Too avoid a potential performance problem it might be useful to raise the goal so that WLM has a better objective to manage to should contention occur. By looking at the velocity distribution, we can see that a goal of 50% or 55% (or even higher depending on how important the work is),would reflect more accurately the demands made by the work in this service class. By adjusting this value, WLM can now ensure that the necessary resources are available to this service class, even during constraint periods, to achieve the desired levels of performance. Should maintaining this level of service be of critical concern, the importance level (currently at 3) could also be raised to increase the priority with which WLM should ensure that goals are met.
If you would prefer, we can run your data through RMF Redux to show you what your reports would look like.
In both of these examples, we have seen two very different kinds of workloads, with each being evaluated based on the data available in the reports. This concentration of information eliminates the need to review voluminous amounts of interval data and spending time with a calculator to arrive at the same conclusions. Using additional information, such as the GOALS SAMPLED field, will indicate whether the data being reviewed is truly representative or that more samples should be included.
The larger the sample set used, which could be several days or weeks, the better the overall information will represent the true behavior of your system.
While it is certainly important to use real-time monitors and data to act on existing problems and current events, these tools are unsuitable for providing the necessary overview to determine how well work is progressing in a general sense. Using this tool to establish a good working policy may help to minimize the variations that result in the day to day problems many organizations spend their time on.
This product represents the beginning of a series of tools to provide greater insight into WLM behavior and management. Future versions will include the ability to format and analyze the SMF Type 99 records, as well as provide greater insight into work elements, such as enclaves and servers to determine how well these components are performing.
For additional information please email Tim Noca or call him at 800-373-9188x112.