07.02.2013 Views

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

Best Practices for SAP BI using DB2 9 for z/OS - IBM Redbooks

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

If transfer time takes the most elapsed time<br />

For this condition, use transaction SM50 to determine whether there is resource<br />

constraint. Check whether there are enough work processes, especially <strong>for</strong><br />

loading to PSA and data targets in parallel. Parallel processing can compete <strong>for</strong><br />

resources. Use the ST06 transaction to determine whether the bottleneck is in<br />

CPU, memory, or network.<br />

For parallel processing (load balancing) to multiple BW application servers,<br />

synchronization among them is necessary because during large data load,<br />

swapping of buffers of master data can occur on a BW application server.<br />

Synchronization overhead of buffers among them can impact per<strong>for</strong>mance. You<br />

can use ST05 to check the SQL trace <strong>for</strong> log table activities. If this problem<br />

occurs, you can try to shut down other application servers and set<br />

rdisp/bufrefmode=sendoff,exeauto during the load phase.<br />

If upload (update) to a PSA takes the most elapsed time<br />

Potential root causes <strong>for</strong> this condition could be I/O contention or the partitioning<br />

configuration. I/O contention is a possible result of high volumes of insertion<br />

during large data loads. Check <strong>for</strong> database I/O, disk layout, and striping<br />

configurations to see whether you can improve it.<br />

The partitioning configuration has an impact as well. If partitions are defined too<br />

large, there may be no parallel database subprocesses used. If partitions are<br />

defined too small, there may be too many parallel database subprocesses used.<br />

Packet size and partition size can be done from transaction RSCUSTV6.<br />

Note that in <strong>DB2</strong> V8, the partition size defined in RSCUSTV6 will not influence<br />

the number of partitions. <strong>SAP</strong> BW table makes use of the <strong>DB2</strong> V8 partition<br />

addition feature (<strong>for</strong> more details refer to 5.1.2, “Management of PSA” on<br />

page 68).<br />

The PSA table partition is automatically created <strong>for</strong> each data request in an<br />

InfoPackage. If multiple data requests on the same Infopackage are loaded to<br />

PSA, then the PSA table partition will have additional table partitions per the<br />

number of loads of data requests. If the same data request load is repeated, a<br />

new partition will be added.<br />

There<strong>for</strong>e, to influence parallelism, split large loads into multiple Infopackages<br />

that each have only one data request. For more in<strong>for</strong>mation refer to <strong>SAP</strong> note<br />

number 485878 <strong>for</strong> “<strong>DB2</strong>/390: BW Partitioning the PSA tables”.<br />

Transfer rules and update rules takes too long<br />

For this condition you can use the simulation tool to debug the problem. For<br />

example, in the RSMO detailed window, under Processing, select Data Package<br />

Chapter 8. Query and Load per<strong>for</strong>mance 155

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!