I've opened a support call with HP/VMware on this issue and while I'm waiting on their response, I thought I'd try the community. We've recently started virtualizing some of our more demanding Oracle environments and our DBAs reported that the latest VM we deployed (Oracle Linux 6.4) doesn't have HugePages enabled. I would think this setting would need to be done at the Guest rather than the Host, but am not entirely sure. Thank you in advance for any insight you can offer.
Please post the output of the following commands from the guest OS:
# cat /proc/cpuinfo
# cat /proc/meminfo
# cat /sys/kernel/mm/transparent_hugepage/enabled
# numactl --hardware
# numactl --show
Does the VM have more than 8 vCPUs? I'm not sure but I think large memory pages might require a NUMA topology on the guest OS. By default ESXi only presents a virtual NUMA topology to the guest if the VM is configured for more than 8 vCPUs.
Huge pages has nothing to do with NUMA, regardless of whether vNUMA is enabled (by default on vcpu > 😎 or not, huge pages is a Guest OS thingy.
nr.hugepages needs to be setup in /etc/sysct.conf along with use_large_page parameter in Oracle init/spfile
Sorry to hijack this thread but I have this exact problem.
I am running ESXi 6.7 and I have enabled hugapages support as per Huge Pages
but this still doesn't do anything
cat /proc/meminfo | grep -i huge
AnonHugePages: 329728 kB
ShmemHugePages: 0 kB
Hugepagesize: 2048 kB
This should be to 1G
The other parameters seem to go well:
Can someone help ? as my container doesn't start because of this ....
[ 496.326645] hugetlbfs: Unsupported page size 1024 MB
[ 497.030697] hugetlbfs: Unsupported page size 1024 MB
[ 497.632851] hugetlbfs: Unsupported page size 1024 MB
[ 498.237018] hugetlbfs: Unsupported page size 1024 MB