Users lured by the scalability promise of Non-Uniform Memory Access (NUMA) architectures, but wary of its relative newness in commercial environments, may find some reassurance in a recent announcement from Sequent Computer Systems.
The vendor is adding features that give its NUMA servers, which first started shipping seven months ago, greater reliability, increased memory and more storage space.
NUMA technology is a performance-enhancing way of tying together small groups of processors into one large cluster. For example, a 16-processor server can be arranged into a cluster of four nodes with four processors each.
The technique allows vendors to extend ser- ver scalability beyond the limits imposed by symmetrical multi-processing (SMP) architectures. Although most SMP servers stop scaling effectively after 16 processors, Sequent's NUMA servers now handle up to 32 Intel processors, and eventually will scale up to 252 processors.
By bringing high-availability clustering to NUMA, Sequent gives users a way to increase the reliability of their applications, observers said. The company has introduced a fibre-channel-based interconnect technology that lets users link two NUMA servers in a high-availability cluster. In that setup, if one server goes down, the other automatically takes over.
The servers can be spaced as much as 10km from one another.
Sequent also announced support for up to 16Gb of Very Large Memory support on its NUMA servers. That kind of memory capacity lets corporations run large applications - such as entire databases - directly on memory.