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.

2.1.5 Real storage<br />

<strong>DB2</strong> V8 extends upon its improvements in storage management and has also increased<br />

some defaults. All of these changes do not eliminate completely short-on-storage conditions.<br />

You still need to proactively monitor your storage utilization; keep running your <strong>DB2</strong> PE<br />

storage statistics reports and watch your storage utilization over time.<br />

You can experience some real storage growth usage when you migrate from <strong>DB2</strong> V7 to V8.<br />

The growth is due to a number of factors, the most important is 64-bit addressability, Unicode,<br />

long names support and long key support. The 64-bit instructions are inherently longer than<br />

31-bit counterparts; this effect can increase the size of control blocks that store pointers by a<br />

factor of two. Longer names require larger definitions and space.<br />

You need to make sure that the buffer pools are 100% backed by real storage. <strong>DB2</strong> uses an<br />

LRU or FIFO scheme to reclaim pages. As a result, if the oldest pages are paged out to disk,<br />

the paging impact on the system would be significant. In the zSeries architecture, if the buffer<br />

pool is not backed by main storage, the pages would be stored in and out of auxiliary storage<br />

(that is, disks), since there is no configurable expanded storage to provide some hierarchy of<br />

buffering. Expanded storage is not exploited by z/<strong>OS</strong> in the z/Architecture® modeI. It is wise<br />

to prevent excessive paging to auxiliary storage. Use RFM Monitor II and III to monitor the<br />

real storage frames and auxiliary frames used by the <strong>DB2</strong> address spaces. It is very<br />

important to keep in mind that your <strong>DB2</strong> subsystem should not page.<br />

Important: Ensure the buffer pools are backed up by sufficient real storage.<br />

2.1.6 Compatibility mode<br />

Once you commit to migrating to <strong>DB2</strong> V8, the first stop you make is compatibility mode (CM).<br />

As soon as you are in CM, there is no new function. Well, there is actually some new function;<br />

to start with, <strong>DB2</strong> is running in 64-bit mode already, however, there is no new function which<br />

can preclude a fall back to V7. It is not officially documented what is available in CM, and it is<br />

subject to change with maintenance. During this phase, you can test the functionality of<br />

existing applications to ensure they run without problems. Only during this phase can you<br />

revert back to <strong>DB2</strong> V7, so once you make the commitment to proceed to the next two stops,<br />

enabling-new-function mode (ENFM) and new-function mode (NFM), there is no turning back.<br />

You will probably remain in CM <strong>for</strong> a short period of time until you are com<strong>for</strong>table that it is<br />

safe to move <strong>for</strong>ward. If you have several <strong>DB2</strong>-interconnected subsystems, DRDA or<br />

otherwise, migrate all systems to CM be<strong>for</strong>e migrating any system to NFM.<br />

Keep in mind that transition among the modes in a data sharing group is an instantaneous<br />

group wide event; it is not possible to have some members in a data sharing group in one<br />

mode while others are in a different mode. All members of a data sharing group are in the<br />

same mode.<br />

Note: Customers typically stay in CM mode <strong>for</strong> a period of several weeks to a few months.<br />

What we have measured in CM:<br />

► The measurements of the CPU time per commit <strong>for</strong> non-distributed IRWW workload have<br />

shown a slight increase of 4% in accounting CPU time between <strong>DB2</strong> V7 and <strong>DB2</strong> V8 CM.<br />

The IRWW workload is described in <strong>DB2</strong> <strong>for</strong> MVS/ESA <strong>Version</strong> 4 Data Sharing<br />

Per<strong>for</strong>mance <strong>Topics</strong>, SG24-4611. However, there is a considerable decrease in total CPU<br />

time <strong>for</strong> the DBM1 address space (about 20%).<br />

► Other workloads have a different behavior.<br />

Chapter 2. Per<strong>for</strong>mance overview 13

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

Saved successfully!

Ooh no, something went wrong!