Technical Blog Post
Abstract
Case Study : Db2 Reorg takes longer in GDPC environment as opposed to a regular purescale environment
Body
Came across a behavior where Db2 Reorg was taking atleast 3X times longer in GDPC environment as opposed to a regular purescale environment.
We gathered various performance monitoring and diagnostic data (eg db2mon reports, db2trc etc) and reviewed the data /repro in lab. From the db2mon data, we can see that WAR (writing a new version of the page, which goes to both cf's) and RAR (read and register page) calls from a member to the CF on the other site take longer in a GDPC env.
For example, the avg time taken for WriteAndRegister (Write and Register command) from member 0 to cf 129 is higher than from member 0 to cf 128. And vice versa for WriteAndRegister calls for member 1. In a regular Purescale env, the average times for these calls from a member to both cf will be relatively closer.
Also, in a GPDC env, each IO to a filesystem will perform IO to both replicas. This will result in some overhead for write activity in the GDPC env as opposed to the regular purescale env.
The above factors will affect the time taken to complete a reorg operation in a GDPC env as opposed to a regular purescale env.
Given the metrics below, this behavior regarding the time was working as per design.
===================================================================
Round-trip CF command execution counts and average response times
===================================================================
db2mon.10230845 :
MEMBER ID CF_CMD_NAME TOTAL_CF_REQUESTS AVG_CF_REQUEST_TIME_MICRO
------ ------ ------------------------------ -------------------- -------------------------
0 128 WriteAndRegister 1128 21.60
0 129 WriteAndRegister 1128 739.55
1 128 WriteAndRegister 12883 747.36
1 129 WriteAndRegister 12884 751.68
0 128 ReadAndRegister 1770 121.51
0 129 ReadAndRegister 1791 1291.98
1 128 ReadAndRegister 1813 766.60
1 129 ReadAndRegister 3021 652.21
0 129 ReadAndRegisterMultiple 26 17559.50
1 129 ReadAndRegisterMultiple 18353 4852.85
Thanks,
Shashank Kharche
IBM DB2 LUW Lab
UID
ibm13285927