Memory Leak in MemToLeave (Non-Buffer Pool)
Does anybody know a good way to identify what is fragmenting my MemToLeave area?
I am using LiteSpeed for SQL 2005 on an SQL Server 2000 Enterprise SP3a default installation running on Windows 2003 Enterprise SP2. Total RAM is 8 GB. I am using AWE, PAE, and 3GB. The min and max server memory are fixed at 6757 MB. The startup parameter -g is 640.
After several days, my large database (> 100 GB) start to fail with a SQL Server ErrorLog message of "Warning: Failed to reserve contiguous memory of size= 65536". Eventually, other application logins start to fail. Then I have to reboot, and everything works again.
I just discovered an extended stored procedure (xp_memory_size) that LiteSpeed installs to show the largest contiguous block of MemToLeave. This is great. At least now I can see the failure coming.
Testing a LiteSpeed backup, I can see it uses the MemToLeave area, which drops in available contiguous size as expected, but returns to original size after the backup completes as also expected.
However, something else besides LiteSpeed has grabbed or spintered the MemToLeave. It was 306 MB when I first ran xp_memory_size today. Now it is 266 MB.
One last thing. Each server reboot, I get a SQL Server ErrorLog message of "initdata: Warning: Could not set working set size to 2724864 KB". This may be related, but I'm not sure exactly how. I'm probably trying to squeeze out too much memory.
All ideas and suggestions will be appreciated.
Thanks.
