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.

Req Out<br />

(req / sec)<br />

CHPID Busy<br />

(%)<br />

We can see from Table 8-8 that Protocol Level 2 generates a lot less XCF messaging traffic to<br />

and from the coupling facility.<br />

Table 8-9 shows extracts from RMF CF activity reports <strong>for</strong> the test workload.<br />

Table 8-9 Protocol Level 2 - RMF CF Activity Report extract<br />

Lock Structure<br />

Activity<br />

Total Requests<br />

(/ sec)<br />

Deferred Requests<br />

(/ sec)<br />

Contention<br />

(/ sec)<br />

False Contention<br />

(/ sec)<br />

CF Utilization<br />

(%)<br />

<strong>DB2</strong> V7 <strong>DB2</strong> V8<br />

2,300 2,200 17 18<br />

97.43 86.53 22.27 12.74<br />

<strong>DB2</strong> V7 <strong>DB2</strong> V8<br />

8,418.33 13,541.67<br />

790.00 6.92<br />

790.00 6.89<br />

115.00 2.58<br />

5.4 6.9<br />

This table is rather interesting. It shows many more requests to the lock structure as a result<br />

of using Protocol Level 2. This increase is in synchronous requests, as the number of<br />

deferred or asynchronous requests have decreased.<br />

Under <strong>DB2</strong> V7, the IX L-locks are mapped to X-XES locks and there<strong>for</strong>e would occupy a lock<br />

table entry in the lock structure. Once an XES essentially “owns” a lock table entry, by owning<br />

the X-XES lock, that XES becomes the global lock manager <strong>for</strong> the lock table entry and has<br />

the responsibility to resolve any contention that exists, (false, XES or real contention), <strong>for</strong> that<br />

entry.<br />

Once potential contention has been identified by two lock requests hashing to the same lock<br />

table entry, the global lock manager XES must resolve the contention. The other XESs in the<br />

group are instructed to send the lock in<strong>for</strong>mation they have that pertains to the lock table<br />

entry to the global lock manager XES so it can determine whether the contention is real or<br />

false. Resolution is possible as each XES holds in<strong>for</strong>mation about locks that its IRLM has<br />

passed to it. Some optimization is in place so that once an XES “owns” a lock table entry,<br />

other XESs in the group do not have to interact with the lock structure in the coupling facility.<br />

Under <strong>DB2</strong> V8 the IX L-locks are mapped to S-XES locks and do not occupy a lock table<br />

entry. The end result is there will be fewer lock table entries “owned” by X-XES locks. The<br />

good side is that this reduces XES contention (IX-L locks are now mapped to S-XES locks<br />

which are compatible) and reduce false contention (S-XES locks do not “own” lock table<br />

entries, so fewer lock table entries are occupied). The down side is that more lock requests<br />

must be propagated to the coupling facility to determine if any contention may exist. The<br />

majority of these requests are synchronous requests as contention does not exist (the<br />

hashed lock table entry is not in use) and the request does not need to be suspended to<br />

Chapter 8. Data sharing enhancements 329

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

Saved successfully!

Ooh no, something went wrong!