You are currently on IBM Systems Media’s archival website. Click here to view our new website.

MAINFRAME > Administrator > Db2

Making DB2 Zip With IBM’s zHyperWrite


A Client Example

The client example depicted in Figure 1 shows the improvements seen by one of the first adopters of zHyperWrite. This large financial institution ran an experiment to make 2.7 million database updates with a commit made after every three updates. DB2 elapsed time improved by 28 percent. The same client observed a decrease in CPU time due to a reduction in the commit time. This also reduced the lock hold time, reducing lock contention wait time. These improvements in CPU time can offset the small CPU cost for parallel execution of primary and secondary write I/O operations.

The results of an IBM internal performance test for the zHyperWrite technology is shown in Figure 2. At short distances between primary and secondary storage subsystems, performance testing showed a 300 microsecond component of I/O service time. If the secondary control unit was more distant, I/O service time for synchronous replication increased about 10 microseconds per kilometer—the round trip speed of light through optical fiber. With the zHyperWrite feature enabled, 40 percent reduction was achieved in the DB2 log write latency and corresponding commit response time.

Figure 3 shows experimental results for DB2 log writes of different sizes with zHyperWrite turned off and on. With High Performance FICON, DB2 log write latency was improved by up to 43 percent, resulting in a 179 percent increase in DB2 log write throughput.

How It Works

Existing configuration definitions specified using GDPS* or TotalStorage Productivity Center for Replication HyperSwap are used for zHyperWrite. The same internal tables are used in z/OS to manage HyperSwap and zHyperWrite. All that’s needed to take advantage of this capability—besides installing the code for z/OS, DB2 and DS8870 firmware—is to enable the function by database inside the DB2 customization. In other words, set the on switch.

Eligible I/O operations are automatically identified by the middleware (e.g., DB2) and processed accordingly. Without zHyperWrite, storage subsystems are responsible for replicating the data to secondary control units. This serial processing can add up to 300 microseconds to the I/O service time (see Figure 4).

Figure 5 shows the flow of data for write operations eligible for zHyperWrite. In this case, z/OS will execute multiple parallel write requests, eliminating delays imposed by serial processing for control unit-based synchronous replication. Write operations are posted complete to applications and middleware when all have completed. If an I/O operation ends in an error, it is retried without zHyperWrite. As always, permanent errors result in a HyperSwap event, allowing the system to continue without disruption. If a site disaster requires a DB2 restart, databases can be recovered from primary devices or any of the secondary devices. DB2 recovery will apply the log to make the database consistent to the point of time in the DB2 log.

It is important to emphasize that zHyperWrite is managed on an I/O operation case-by-case basis. Both eligible and ineligible data for zHyperWrite acceleration can reside on the same devices concurrently. This hybrid approach is a flexible way to manage I/O configurations.

New Resource Management Facility instrumentation reports on I/O measurements for the Metro Mirror secondary devices to identify any hot spots or bottlenecks in the I/O configuration. Additionally, new instrumentation and data added to SMF 42-6 records allows an account for additional I/O requests generated by the system.

Continued Leadership

The z Systems* platform has a strong tradition of close integration of its I/O capabilities throughout the I/O stack including the applications, middleware, OS, channel subsystem, I/O channels, I/O fabric and devices. This integration allows innovative optimizations that improve performance, scale and resilience for the critical workloads that run on the mainframe. The zHyperWrite technology is the latest innovation from IBM storage, DB2 and z/OS to deliver value for z Systems clients.

Calline Sanchez is the VP of IBM Systems Storage program management.

Harry M. Yudenfriend is an IBM Fellow with Systems and Technology Group, System z and Power who joined IBM in 1980. He was named an IBM Master Inventor in 2001 and has achieved his 33rd invention plateau.

Peter Sutton is an IBM Distinguished Engineer for IBM Z core technology.



Advertisement

Advertisement

2019 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.

An Intuitive Approach to DB2 for z/OS SQL Query Tuning

Real-world samples provide valuable lessons.

MAINFRAME > ADMINISTRATOR > DB2

A Db2 Utilities Migration Project

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters