Bookmark and Share
RSS

Recent Posts

Group Buffer Pool-Related Enhancements in DB2 11

November 03, 2015

Several enhancements in DB2 11 are designed to ease and eliminate storage constraints on group buffer pools and to reduce elapse time processing in applications.

The first enhancement involves group buffer pool castout processing. It's worth mentioning that castout is the process of writing pages from the group buffer pool to disk in a data sharing environment. A group buffer pool is a DB2 for z/OS coupling facility cache structure that a data sharing group uses to cache data and ensure the data is consistent for all members. (Note: Here's a complete list of DB2 for z/OS terms.)

One of the problems in data sharing environments with heavy write activity (i.e., applications performing INSERT, UPDATE, DELETE) is that pages are written to the group buffer pool faster than the castout engines can process them. Such inefficient castout processing often leads to application response time issues. In extreme cases, it could cause a group buffer pool full condition.

DB2 11 is designed to make castout processing more efficient through these enhancements:
  • Overlapped castout and write I/O, resulting in reduced wait time for I/O completion.
  • Simplified castout notification, resulting in smaller notify message size sent to castout owners.
  • More granular class castout threshold. This is accomplished through the capability to specify CLASST castout threshold as a number of pages rather than as a percentage.
Each of these enhancements function in DB2 11 conversion mode (CM), meaning that no user action is required to enable them, and there's no need to move to new function mode (NFM). According to IBM, customers could see a 90 percent or greater reduction in average wait times to update commit processing as a result of these changes. In addition, IBM has eliminated Write Failed – No Storage and Write to Sec-GBP Failed conditions in situations that previously received these messages. This is another indication of efficient castout processing.

Another feature implemented with DB2 11 is the group buffer pool write around protocol. This is designed to help manage large write activity and avoid storage shortage problems that typically occur with utility or batch jobs. Rather than write pages directly to the group buffer pool, DB2 instead writes pages directly to DASD. Note that group buffer pool write around is tied to two thresholds with fixed values (i.e., they cannot be changed):
  • Group buffer pool threshold (GBPOOLT) is set at 50 percent.
  • Class castout threshold (CLASST) is set at 20 percent.
When GBPOOLT hits 50 percent -- meaning 50 percent of the group buffer pool is occupied with changed pages -- write-around is invoked for all objects. This continues until GBPOOLT drops to 40 percent.

When CLASST hits 20 percent -- meaning 20 percent of a class castout queue is occupied by changed pages -- write-around is invoked for the page set/partition. This continues until the CLASST drops to 10 percent.

To determine if pages have been written to DASD, issue the -DISPLAY GROUPBUFFERPOOL command with the MDETAIL option. In the output from this command, look at the WRITE-AROUND PAGES field in message DSNB777I.

These group buffer pool storage and performance enhancements offer another good reason to consider an upgrade to DB2 11.

Posted November 03, 2015 | Permalink

comments powered by Disqus