For MQ sites looking for ways to improve performance, the recent support of zHyperWrite will likely come as welcome news. Recent versions of Db2 and MQ have provided support for zHyperWrite with the goal of improving overall subsystem performance. zHyperWrite improves response times for log writes in shops that use synchronous replication to ensure a copy of the log exists on a remote secondary device.
In this blog we’ll examine the before-and-after measurements of a recent zHyperWrite implementation for MQ logging in a large z/OS environment.
Importance of MQ Log Performance
Speed of logging is very important to the health and performance of the overall subsystem. For MQ, improved log performance improves scalability and the ability to handle spikes in workload. It also improves the overall efficiency and lowers the cost of operation.
Before zHyperWrite, I/O drivers such as Media Manager would issue a write to the primary device, and the primary would initiate the mirroring to the secondary in a peer-to-peer manner.
With zHyperWrite, the mirroring is initiated concurrently with the primary I/O, speeding up the process. Note that this results in the number of reported writes doubling, since Media Manager is writing to both the primary and secondary device, and the peer-to-peer activity is eliminated. The improvement in response time is greater with short distances between primary and secondary.
Analyzing MQ Log Performance Before and After Implementing zHyperWrite
Figure 1 below shows an example of implementing zHyperWrite in a large MQ environment, with comparison data for key before-and-after metrics.
Figure 1 shows MQ logging data for several business days before the zHyperWrite implementation on the left and several business days after implementation, on the right. You can see that the I/O rate doubled, since there are now FICON channel I/Os to both the primary and secondary device. Metro Mirror ops/second are shown in red. These are the ops that were peer to peer from primary to secondary but are now direct writes from the system to the secondary. Notably, the response time was greatly improved, roughly cut in half.
Breaking down the response time components, as shown in Figure 2 below, we can see that the improvement is primarily due to the elimination of disconnect time (the red area).
Further breaking down the components of disconnect time shows that the improvement comes from eliminating the synchronous copy contribution. This data can be obtained from the type 74 records combined with intelligence built into IntelliMagic Vision. Here is an example of a volume used for MQ logging, not using zHyperWrite.
Note that the overall disconnect time is approximately .2 milliseconds, as we saw at the dataset level in Figure 2 before zHyperWrite. Also note that almost all the disconnect time is the synchronous copy contribution (in red).
In contrast, the Figure 4 shows a volume with MQ logging using zHyperWrite, with essentially zero disconnect time.
The most important bottom-line comparison is the overall response time improvement. Figure 5 shows a 54% improvement for this heavily used MQ subsystem.
zHyperWrite Implementations Receiving Positive Response Time Results
Thus far, the customers we have seen implementing zHyperWrite are receiving positive results. The customer site referenced above is already planning further rollout to additional subsystems and environments.
Of course, in any complex implementation project, it is important to be able to clearly communicate the initial results as well as further improvement as you proceed with additional stages of implementation. Quantifying the benefit of one project can pave the way for approval of the next. Having the reporting you need, in an easy to use and visually compelling manner, makes it much easier.
If you are considering implementing zHyperWrite, or any other new technology, at your site but want to be sure you have a clear baseline of performance results to measure the impact, reach out to us to help you set that up.
You May Also Be Interested In:
From Taped Walls to Your PC: z/OS Configuration Made Simple with Topology | IntelliMagic zAcademy
In this webinar, we explore a technique that integrates diverse data sources from z/OS, presenting them in a clear, concise, and interactive visual format.
Banco do Brasil Ensures Availability for Billions of Daily Transactions with IntelliMagic Vision
Discover how Banco do Brasil enhanced its performance and capacity management with IntelliMagic Vision, proactively avoiding disruptions and improving cross-team collaboration.
Unraveling the z16: Understanding the Virtual Cache Architecture and Real-World Performance | IntelliMagic zAcademy
This webinar will provide detailed z16 insights, including an overview of all key z16 advances and a deep dive into the new cache architecture.