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.

Attention: The ranges of percentages reported in the row DBM1 VS (Virtual Storage) in<br />

MB below 2 GB actually reflect the percent increase moving from V7 to V8 with and<br />

without the virtual buffer pools. Naturally, a decrease in the virtual storage is possible since<br />

the virtual buffer pools move above the 2 GB bar. In the case of large buffer pools, they are<br />

likely to be using hiperpools and data space buffer pools, so this would not apply. You also<br />

see a larger decrease if you compress some of your data, since compression dictionaries<br />

also move above the bar.<br />

We can make the following conclusions about the per<strong>for</strong>mance data in Table 4-6:<br />

A general increase in <strong>DB2</strong> V8 DBM1 virtual storage below the 2 GB bar in the distributed<br />

environment is observed:<br />

► The virtual storage usage <strong>for</strong> the DBM1 address space below the 2 GB bar in V8<br />

increases in the range of 24 to 29%. This increase does not take into account the storage<br />

<strong>for</strong> virtual buffer pools. If it did, then the virtual storage usage <strong>for</strong> the DBM1 address space<br />

below the 2 GB bar in V8 decreases in the range of 55 to 48%.<br />

This is primarily due to larger thread-related storage structures required by <strong>DB2</strong> V8 to<br />

support new functionality such as long names, longer SQL statements, Unicode and 64-bit<br />

addressing.<br />

The DBM1 virtual storage increase per distributed user thread is in the range of 49 to 72%.<br />

The predominant contributors are non-system thread storage and user stack storage.<br />

For the SQLJ workload the virtual storage per active connection increased by 201%.<br />

► Total DBM1 VS without BP - up to 38% increase <strong>for</strong> the embedded SQL workload<br />

► Per User Thread - up to 73% increase <strong>for</strong> the SQLJ workload<br />

► Per System Agent - up to 76% increase <strong>for</strong> the CLI workload<br />

As virtual storage increases, real storage usage will generally increase in order to back the<br />

virtual storage requirement.<br />

User thread storage comparison<br />

The increase in storage is primarily thread-related, as the GETMAINed areas and fixed<br />

storage pools did not change with the different workloads in either V7 or V8.<br />

<strong>DB2</strong> V8 uses more storage <strong>for</strong> each thread in the DBM1 address space due to:<br />

► Thread structures increased in size to support larger names in <strong>DB2</strong>.<br />

► The RDS OP pool storage that was used at bind time in V7 is now acquired from the<br />

thread's variable storage pool.<br />

► Some <strong>DB2</strong> structures must increase in size to support 64-bit addresses.<br />

This section compares the user thread storage in DBM1 below the 2 GB bar and references<br />

Figure 4-5.<br />

Chapter 4. <strong>DB2</strong> subsystem per<strong>for</strong>mance 151

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

Saved successfully!

Ooh no, something went wrong!