We recently upgraded one of our systems to a P5, and upgraded AIX 5.2
from ML06 to TL09. We moved the raid enclosure and drives over to the
new box, so other than the new adapter card, we have the same I/O
system. We do weekly maintenance which consists of doing an export,
followed by a load, of the data back into the table, essentially doing a
reorg, but allowing for stopping in between if necessary.



The issue we are having is that the length of time to complete the
export/load combination has gone from 5 hours, to more than 20 hours,
with no change in data volume. Working with IBM, they determined that
the additional time was being spent in Memory Mapped I/O, and had us
turn MMAP off during the maintenance process (DB2_MMAP_READ=OFF,
DB2_MMAP_WRTIE=OFF). This partially improved the speed from 20+ hours
down to 10+ hours.



We recently upgraded a second system, with a completely new raid
enclosure and drives, and have had similar results, the maintenance went
from 3 hours, to over 15hours. Making the MMAP changes brought it down
to 6 hours.



It seems to me that there has to be an issue with either the P5 box, or
the AIX TL09, or both, but IBM is denying that there is an issue. Has
anyone else had a similar experience?



Michael Barton

Application Analyst - Level 2

Bally Technologies




DB2 & UDB email list listserv db2-l LazyDBA home page