19.06.2013 Views

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

DB2 UDB for z/OS Version 8 Performance Topics - IBM Redbooks

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

and secondary GBP structures have the CF request batching capability. Since CF request<br />

batching will only be used <strong>for</strong> reading castout pages from the primary GBP, there is no<br />

additional CPU saving <strong>for</strong> castout with or without GBP Duplexing. However, if you do not<br />

consider prefetch, then the additional CPU saving with CF request batching in a duplexed<br />

GBP environment will be half of the CPU saving without GBP duplexing.<br />

Table 8-4 shows extracts from RMF reports <strong>for</strong> the same batch workload.<br />

Table 8-4 CF Request Batching - RMF extract (batch)<br />

Once again, we can see z/<strong>OS</strong> 1.4 results in significantly fewer requests to the coupling facility<br />

(6,715 compared to 14,627). We can also see the service times <strong>for</strong> both synchronous<br />

requests and asynchronous requests to the coupling facility have increased slightly, as well as<br />

notice the coupling facility utilization has decreased slightly.<br />

The table also shows quite a jump in asynchronous requests when CF request batching was<br />

used, over 12% of all requests, compared with only 2% when CF request batching was not<br />

used. This trend is similar to what we noticed when we earlier reviewed the RMF extracts <strong>for</strong><br />

the OLTP workload. The increase can be attributed to the heuristic conversion by the CFCC<br />

of synchronous requests to asynchronous requests due to more work being processed in<br />

each batch request.<br />

Table 8-5 compares the CPU used by the DBM1 address space <strong>for</strong> the OLTP workload<br />

compared with the batch workload.<br />

Table 8-5 CF Request Batching - <strong>DB2</strong> CPU<br />

This table shows CF request batching has potential <strong>for</strong> significant saving in CPU charged to<br />

DBM1 address space, in both OLTP and batch workloads, but particularly batch.<br />

The overall CPU times in this table include the transaction class 2 CPU times, the MSTR,<br />

DBM1 and IRLM CPU times. So, we can see the overall impact of CF request batching was<br />

324 <strong>DB2</strong> <strong>UDB</strong> <strong>for</strong> z/<strong>OS</strong> <strong>Version</strong> 8 Per<strong>for</strong>mance <strong>Topics</strong><br />

z/<strong>OS</strong> 1.3 z/<strong>OS</strong> 1.4<br />

Requests / sec 14,627 6,715<br />

Sync Requests<br />

Serv time<br />

(usec)<br />

Async Requests<br />

Serv time<br />

(usec)<br />

% of all req 10.5 98.0 12.0 87.7<br />

% of all req 209.8 2.0 241.4 12.3<br />

CF Utilization (%) 15.7 14.4<br />

DBM1 - (OLTP)<br />

(msec / commit)<br />

z/<strong>OS</strong> 1.3 z/<strong>OS</strong> 1.4 Delta<br />

(1.4 / 1.3)<br />

0.536 0.545 0.427 0.473 -17%<br />

Overall CPU time 3.229 3.224 3.111 3.368 +0%<br />

DBM1 - (Batch)<br />

(msec / commit)<br />

119.88 130.11 81.5 89.68 -32%<br />

Overall CPU time 656.40 709.74 618.98 680.89 -5%

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

Saved successfully!

Ooh no, something went wrong!