Memory issue on SLES 11

gtyrrell · 03-23-2014, 05:21 PM

I've got data warehouse software (IQ 16 SP03) that's running into a memory allocation issue while doing table loads on SUSE Linux Enterprise Server 11 sp3, but not on other operating systems. IQ has no known issues along these lines, and a bunch of IQ experts have agreed the issue is at the O/S. Unfortunately, I don't have the option of delivering on another operating system.

IQ is correctly configured to accept sufficient memory for what we're trying to do, but we're getting errors every time we try to grab more than 1GB of memory during loads:

Code:

Error 1013130 "All IQ large memory has been used, allocation canceled [size: 1073741887]"

Somebody more Linux-conversant had me run a stack trace during the load and he says he can see that he can "see the messages about Exception in strace output, however subsequent memory mapping does not throw an error", but he hasn't yet come up with a reason why this is happening. I was hoping somebody here might have an idea of what the stack trace is saying, and what might fix the situation.

Command used to produce the stack trace:

Code:

  strace -f -ttt -T -o OUTPUT_FILE -p IQ_SERVER_PID

Trace output (started just prior to load attempt, ended just after failure) attached.

Thanks and appreciation in advance.

Gary
gary.tyrrell@sap.com

syg00 · 03-24-2014, 12:54 AM

Being SLES, I would think you should be asking this question of SuSE.
What is meant by "large memory" in the message above - is the software attempting to allocate huge/large pages, and/or shm ?. Let's see the output from the following command issued in a terminal.

Code:

cat /proc/meminfo

gtyrrell · 03-24-2014, 04:57 PM

"Large memory" is an IQ designation for one of three major allocations inside its memory pool; it's not attempting huge/large pages (IQ doesn't support them), but it is attempting to allocate from shared memory. The configuration setting that limits how much large memory can be requested is 2048MB by default, and changes to that setting don't affect the error, which always occurs at 1GB (even when "large memory" is told to limit to as little as 300MB)

Output of cat /proc/meminfo:

Code:

MemTotal:        8194320 kB
MemFree:          898844 kB
Buffers:          326440 kB
Cached:          6107596 kB
SwapCached:           24 kB
Active:          2123488 kB
Inactive:        4654904 kB
Active(anon):     218556 kB
Inactive(anon):   125904 kB
Active(file):    1904932 kB
Inactive(file):  4529000 kB
Unevictable:           0 kB
Mlocked:               0 kB
SwapTotal:      33554428 kB
SwapFree:       33554388 kB
Dirty:                96 kB
Writeback:             0 kB
AnonPages:        344324 kB
Mapped:            48112 kB
Shmem:               112 kB
Slab:             399968 kB
SReclaimable:     367244 kB
SUnreclaim:        32724 kB
KernelStack:        3088 kB
PageTables:         4376 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:    37651588 kB
Committed_AS:    1314392 kB
VmallocTotal:   34359738367 kB
VmallocUsed:      300388 kB
VmallocChunk:   34359433440 kB
HardwareCorrupted:     0 kB
AnonHugePages:    286720 kB
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
DirectMap4k:       49152 kB
DirectMap2M:     8339456 kB

MemFree has been as high as 3.5GB and still showing the same behavior.