We have a bad heap fragmentation situation.
Switching to LFH didn't improve the issue.
0:000> !heap -s
LFH Key : 0x16658011
Heap Flags Reserv Commit Virt Free List UCR Virt Lock Fast
(k) (k) (k) (k) length blocks cont. heap
-----------------------------------------------------------------------------
...
003a0000 00001002 31808 6724 25288 1724 99 124 0 68 LFH
External fragmentation 25 % (99 free blocks)
Virtual address fragmentation 73 % (124 uncommited ranges)
...
We think memory leaks are not the cause (tried umdh.exe and also GFlags
leaks detection). However our application has few growing caches.
Using !heap –flt –h 0x... didn’t help, since we found reasonable figures
for all caches (~10,000 allocation for each cache type).
Are there any good techniques to pinpoint the main fragmentation culprit?
e.g. a dump memory viewer, we can zoom into potential allocations?
or maybe a command similar to !dumpheap
(http://stackoverflow.com/questions/686950/large-object-heap-fragmentation)
Thanks I.A
--
Tal.