It’s a bird… It’s a plane… It’s IMS 15.2!
IMS 15.2, the first continuous delivery release, has arrived, bringing along new key features to streamline your digital transformation journey into the cloud (see the GA announcement letter for more details). But how does its performance hold up?
In this study, we analyze how IMS 15.2 compares to its previous release, IMS 15.1, across several different workloads. The library used to represent IMS 15.1 is the exact same IMS 15 library used in the IMS 15 Performance Evaluation Summary white paper. In addition to new enhancements, the IMS 15.2 library includes all maintenance (APARs) applied since the publication of the IMS 15 performance white paper in 2018.
The IMS Performance team, based out of the IBM Silicon Valley Lab, conducted all performance evaluations on a shared IBM z15. Because of the noise inherent within a shared environment, measurements may have a variability between +/- 1-2%. The following software configuration was used during evaluations:
- z/OS V2.3
- IBM CICS V5.5
- IBM Db2 V12
- Java 8 SR6 64-bit
- IRLM V2.3
- TPNS V3R5
For each evaluation, we compared the External Throughput Rate (ETR),1 Internal Throughput Rate (ITR),2 response time and storage usage while running at 80% CPU busy. Different workloads were utilized to exercise different code paths in IMS. The following types of workloads were used to evaluate the performance of IMS:
- Fast Path (FP)
- Full Function with High Availability Large Database (HALDB)
- Data Sharing Full Function with HALDB and Shared Queues
- Batch Message Processing (BMP)
- CICS-IMS Database Control (DBCTL)
- IMS TM-Db2
- Open Database Manager (ODBM)
- Java Message Processing (JMP)
To drive the online workloads, Systems Network Architecture (SNA) was used during the CICS-DBCTL evaluations while TCP/IP was used for all others.
In our internal evaluations, we found ETR and ITR were about even (within +/- 1-2% noise variability), as seen in Figure 1.
The absolute differences in response times were less than one millisecond across all workloads, as seen in Figure 2.
But we did notice storage increases in both the IMS Control Region (CTL) and the DL/I Region (DLI) for IMS 15.2, as seen in Figure 3.
All workloads showed about a 3% (64-72KB) increase in 24-bit CTL user private. Upon further investigation, the increase in storage was not caused by going from IMS 15.1 to IMS 15.2. Instead, it was caused by an APAR that inadvertently moved the DFSOLC70 module to 24-bit. A fix is currently underway, and the storage increase will be addressed in APARs PH21311 and PH22717.
In workloads accessing OSAM databases, we detected a 24-bit DLI LSQA storage growth of 224 bytes per OSAM data set. Another fix is currently underway, and the storage growth will be addressed in APAR PH23268.
Lastly, all workloads showed a 1% (8KB) increase in 24-bit DLI user private. The increase in storage was expected due to OSAM Media Manager modules being linked into existing DLI load modules residing in 24-bit storage.
Overall, we observed equivalent performance between IMS 15.2 and IMS 15.1. Throughput was comparable and response time differences were minimal. Although we observed increases in storage, the majority of the increases will be fixed in upcoming APARs.
1. ETR is a measure of the actual throughput rate achieved.
2. ITR is a measure of the maximum possible throughput rate that can be achieved with processors at 100% busy.