NUMA (non-uniform memory access) and vNUMA

NUMA (non-uniform memory access) is a method of configuring a cluster of microprocessor in a multiprocessing system so that they can share memory locally, improving performance and the ability of the system to be expanded. NUMA is used in a symmetric multiprocessing ( SMP ) system. An SMP system is a "tightly-coupled,""share everything" system in which multiple processors working under a single operating system access each other's memory over a common bus or "interconnect" path. Ordinarily, a limitation of SMP isthat as microprocessors are added, the shared bus or data path get overloadedand becomes a performance bottleneck. NUMA adds an intermediate level of memoryshared among a few microprocessors so that all data accesses don't have totravel on the main bus.

NUMA can be thought of as a "cluster ina box." The cluster typically consists of four microprocessors (for example, four Pentium microprocessors) interconnected on a local bus (for example, a Peripheral Component Interconnect bus) to a shared memory(called an "L3 cache ") on a single mother board (it could also probably be referred to as a card ). This unit can be added to similar units to form a symmetricmultiprocessing system in which a common SMP bus interconnects all of theclusters. Such a system typically contains from 16 to 256 microprocessors. Toan application program running in an SMP system, all the individual processormemories look like a single memory.

When a processor looks for data at acertain memory address, it first looks in the L1 cache on the microprocessoritself, then on a somewhat larger L1and L2 cache chip nearby, and then on a third level of cache that the NUMAconfiguration provides before seeking the data in the "remote memory"located near the other microprocessors. Each of these clusters is viewed byNUMA as a "node" in the interconnection network. NUMA maintains ahierarchical view of the data on all the nodes.

Data is moved on the bus between theclusters of a NUMA SMP system using scalable coherent interface (SCI)technology. SCI coordinates what is called "cache coherence" orconsistency across the nodes of the multiple clusters.

SMP and NUMA systems are typically used forapplications such as datamining and decisionsupport system in which processing can beparceled out to a number of processors that collectively work on a commondatabase. Sequent, Data General, and NCR are among companies that produce NUMASMP systems.

 

vNUMA (virtual NUMA)

vNUMA (virtual NUMA) is a memory-access optimization method for VMware virtual machines (VMs) that helps preventmemory-bandwidth bottlenecks.

vNUMA (virtual non-uniform memory access)is a memory-access optimization method for VMware virtual machines (VMs) that helps prevent memory-bandwidth bottlenecks. 

The vNUMA hypervisor creates a NUMA topology that closely mirrors the underlying physical server topology, allowing guest operating systems to intelligently access memory and processors. A VM's vNUMA topology will mimic the topology of the host on which it starts; this topology does not adjust if a VM moves to a different host.

Virtual NUMA is especially useful withlarge, high-performance virtual machines or multiple VMs in a vApp. With vNUMAawareness, the VM's memory and processing power are allocated based on theunderlying NUMA topology even when it spans more than one NUMA node.vNUMA-aware VMs must be hardware version 8 or newer and operate on vSphere 5.0or newer, with NUMA-enabled hardware. 

If a VM has more than eight vCPUs, vNUMA isautomatically enabled. The VMware admin can adjust virtual NUMA topology for aVM, or enable vNUMA awareness on a smaller VM, using advanced configurationsettings.VMs using many virtual CPUs (vCPUs) are more likely toaccess more than one NUMA node (local memory tied to a processor) on theunderlying physical server.