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

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

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

<strong>DB2</strong> V8 improves Restart Light. If indoubt units of recovery (UR) exist at the end of restart<br />

recovery, <strong>DB2</strong> now remains running so that the indoubt URs can be resolved. After all the<br />

indoubt URs have been resolved, the <strong>DB2</strong> member that is running in LIGHT(YES) mode will<br />

shut down and can be restarted normally.<br />

A <strong>DB2</strong> member that has been started with LIGHT(YES) that remains active to resolve indoubt<br />

URs will still not accept any new connection requests, except those that originate from<br />

connection names that have indoubt URs.<br />

If DDF is normally active, Restart Light will also start DDF to facilitate the resolution of any<br />

distributed indoubt URs. However, no new DDF connections are allowed. Only resynch<br />

requests will be allowed from <strong>DB2</strong> clients.<br />

This enhancement to Restart Light has the potential to benefit both the availability and<br />

per<strong>for</strong>mance of applications that are still running on other active <strong>DB2</strong> members.<br />

For example, you do not have to bring <strong>DB2</strong> up a second time to resolve any indoubt threads<br />

and remove any retained locks held by those indoubt threads. (Remember that under V7,<br />

<strong>DB2</strong> would terminate immediately after a successful restart, when it is started in light mode.)<br />

Potentially more retained locks are released faster, making the data protected by those<br />

retained locks available sooner <strong>for</strong> applications that are running on other active <strong>DB2</strong><br />

members.<br />

8.5.4 Change to close processing <strong>for</strong> pseudo close<br />

In data sharing, any table space or index which has remained in a read-only state with no<br />

activity <strong>for</strong> a period of time longer than the pseudo-close interval (PCL<strong>OS</strong>ET DSNZPARM) will<br />

be physically closed. This is done in an attempt to reduce data sharing overhead (the object<br />

can become non-GBP-dependent if it is closed on all but one member), but it has the<br />

disadvantage of requiring that the next access to the object must go through dynamic<br />

allocation and data set open again. This can be a significant amount of overhead, particularly<br />

if a large number of objects are accessed at the same time after a period of inactivity (<strong>for</strong><br />

example, at the beginning of the business day).<br />

<strong>DB2</strong> V8 now honors the CL<strong>OS</strong>E YES or NO attribute of the table space or index. The physical<br />

close will now only take place <strong>for</strong> CL<strong>OS</strong>E YES objects, while CL<strong>OS</strong>E NO objects may remain<br />

open indefinitely. This allows you to control the behavior on an object-level basis; <strong>for</strong>merly, the<br />

only means of control was through the pseudo-close DSNZPARMs, and lengthening the<br />

pseudo-close interval to keep data sets open could have other undesirable effects.<br />

There is no per<strong>for</strong>mance impact in a data sharing environment when you specify CL<strong>OS</strong>E<br />

YES. However CL<strong>OS</strong>E NO does have an impact. No really means no now, in a data sharing<br />

environment. If you want things to behave as they did prior to <strong>DB2</strong> V8, (in a data sharing<br />

environment), you need to alter your CL<strong>OS</strong>E NO objects to CL<strong>OS</strong>E YES.<br />

Consider CL<strong>OS</strong>E NO <strong>for</strong> objects that you estimate to be GBP-dependent most of the time and<br />

you do not want the extra open/close overhead associated with CL<strong>OS</strong>E YES. For objects that<br />

are infrequently GBP-dependent, CL<strong>OS</strong>E YES will likely provide better per<strong>for</strong>mance. Once<br />

GBP-dependent, CL<strong>OS</strong>E NO objects will stay GBP-dependent until either DSMAX is hit and<br />

all CL<strong>OS</strong>E YES objects have been closed or until the CL<strong>OS</strong>E NO objects are manually closed<br />

as a result of a <strong>DB2</strong> command or a <strong>DB2</strong> shutdown.<br />

This enhancement is also made available in <strong>DB2</strong> V6 and V7 by PTF UQ74866 <strong>for</strong> APAR<br />

PQ69741.<br />

338 <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>

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

Saved successfully!

Ooh no, something went wrong!