What's all the fuss about NUMA?

IBM is shelling out more than $US800 million for Sequent Computer Systems in no small part to get its hands on Sequent's non-uniform memory access (NUMA) technology.

NUMA is a form of symmetrical multiprocessing (SMP) in which every processor in a server has its own local memory and is able to access other chips' memories as needed to relieve potential bottlenecks. NUMA systems currently scale up to 64 processors and eventually will scale to more than 250, Sequent says.

With traditional SMP systems, shared-memory buses typically cannot accommodate more than 32 processors without bogging down.

NUMA technology was developed in the late 1980s at Stanford University, and Sequent was the first to include it in a product, in the mid-1990s. Sequent and Data General have been NUMA's biggest proponents.

Sequent's customers include Boeing and Ford, and customers typically use NUMA servers to run heavy-duty applications such as enterprise resource planning and business intelligence programs.

Most NUMA machines run on Intel processors, though the technology is not tied to any particular brand of processor. NUMA machines can cost hundreds of thousands of dollars, but Data General claims its servers can cost one-third to one-half less than rival SMP machines. Sequent's NUMA products cost somewhat more than typical SMPmachines.

Other vendors, such as Hewlett-Packard and Amdahl, have dabbled in the NUMA market. In fact, HP has two NUMA models in its server lineup. On the other hand, PC server giants Dell and Compaq have said in the past that they're not considering NUMA in their products.

IBM, meanwhile, already has a dedicated NUMA team in Austin, Texas, and expects big things for the technology as demand for bigger enterprise servers increases.

Join the newsletter!

Error: Please check your email address.

More about AmdahlBoeing AustraliaCompaqData GeneralHewlett-Packard AustraliaIBM AustraliaIntelSequentSequent Computer SystemsStanford University

Show Comments