Do operating systems do anything to account for the non-uniform memory access? Specifically in regards to scheduling or to moving threads to a different core to minimize access time.
The answer is essentially "yes", if the operating system does not employ an intelligent scheduling algorithm, then program performance can vary greatly between runs. Looking around, it seems like there are OSs (specialized for servers) which account for NUMA.
Do operating systems do anything to account for the non-uniform memory access? Specifically in regards to scheduling or to moving threads to a different core to minimize access time.
This comment was marked helpful 0 times.
VMWare explains the dangers and some solutions here: http://pubs.vmware.com/vsphere-4-esx-vcenter/index.jsp?topic=/com.vmware.vsphere.resourcemanagement.doc_40/using_numa_systems_with_esx_esxi/c_what_is_numa.html
The answer is essentially "yes", if the operating system does not employ an intelligent scheduling algorithm, then program performance can vary greatly between runs. Looking around, it seems like there are OSs (specialized for servers) which account for NUMA.
This comment was marked helpful 0 times.