Analyzing and Optimizing System Performance - Single Customer Workshop

The class discusses features available up through z/OS V1.13. This course is targeted for one customer at a time. Before class begins, the customer sends to the instructor RMF (or equivalent) reports, a PARMLIB members listing, and outputs from z/OS display console commands. Based on this input, the lectures are customized to the configuration described in the reports. During the class there will be evaluation, analysis, and discussion about the systems with the technical support staff. A report is developed that includes suggestions to improve overall system performance. This report may include:

• Changes in the WLM policy
• Recommendations for better use of LPAR management
• Modifications in workload execution to reach a better balance
• Suggestions for avoiding wasting resources (such as processors) due to:
  • Inefficient parameters
  • System bottlenecks
  • Inadequate JCL and Parmlib options
  • Potential configuration problems

Here is a list of the major topics covered in the class:

• Processor performance including Parallel Sysplex performance
• I/O considerations
• Central and virtual storage considerations
• Workload analysis

This class is designed to be up-to-date. Each new function in hardware (in a z196, for example) or in software (in z/OS V1R13, for example) affecting system performance is immediately included in the class, with real life examples, if possible. Here are some recent topics:

• WLM Capacity Provisioning
• 1 MB address pages
• HiperDispatch
• MIDAW
• New FICON channel protocol (no more CCWs?)
• Performance recommendations on CF duplexing
• Saving MIPS
• Fact: z/OS is not now a fully preemptible operating system
• What are the consequences of this?
• New discoveries about MSU/hr, MIPS, cycle time, ITR - do not miss
• How to manage the WLM goals of CICS, IMS/DC and WebSphere workloads.
• The enclave concept
• Trading central storage by CPU (such as DB2 buffer pool page fixing) and by I/O (such as VSAM SMB)
• How to encrypt without spending many CPU cycles
• Why we need zAAPs and zIIPs and how they are managed by z/OS
• What is CMR delay during an I/O operation?
• Capping or not capping. If yes, what type?
• The WLM aging algorithm is implemented. Consequences?
• How to improve DASD performance
• The mysterious QDIO algorithm
• The possible revival of IRD dynamic channel management (DCME) for FICON channels