-
Enhancement
-
Resolution: Won't Fix
-
P3
-
None
-
linux
Most container images that we have worked with do not include libnuma.so causing the standard UseNUMA code to be non functional.
It may be possible to take advantage of the OSContainer function to provide some limited NUMA support in os_linux.cpp.
Igor Veresov suggests ...
I think there is a way to do a poor mans NUMA-aware allocator. There is getcpu(2) that can return the numa node that a thread is currently running on. There is no libc wrapper, but the syscall is there since 2.6.19. So, that gives us an implementation of os::numa_get_group_id().
Next, use OSContainer::cpu_cpuset_memory_nodes to implement os::numa_get_groups_num() when running in a container.
- relates to
-
JDK-8241423 NUMA APIs fail to work in dockers due to dependent syscalls are disabled by default
-
- Resolved
-