Next-generation Large Memory HPC Nodes

Commodity class High Performance Computing clusters (HPCC) or Beowulf clusters have served the HPC community well over the last decade. However, in the next 2 years, we will see a new generation of HPCC systems - the Cached Coherent Non-Uniform Memory (ccNUMA) clusters - built from commodity x86 systems with a ccNuma enabled interconnect which will displace the current generation of HPC clusters and large SMP servers.

NumaConnect
NumaScale’s NumaConnect™ technology enables scalable servers at a fraction of the price of current enterprise SMP systems. NumaConnect provides a single system image to a cluster of x86 commodity servers by providing full support for virtualization of processing, memory and IO, and can be used with standard operating systems like Linux, Windows and Solaris. Standard enterprise applications like Exchange or databases like MySQL or Oracle - can run without any code changes. Of course HPC/MPI applications will run without any change except FASTER! Download whitepaper here!

  • Scalable, Directory Based Cache Coherence Protocol 
  • Write-back cache for Remote Data: 2-4-8-(16)GB options, standard SDIMMs
  • ECC protected with background scrubbing of soft errors
  • 16 coherent + 16 non-coherent outstanding memory transactions
  • Support for single-image or multi-image OS partitions
  • 3-way on-chip distributed switching for 1D, 2D or 3D Torus topologies
  • 30GB/s switching capacity per node
  • HTX connected - 6.4GB/s
  • <20W power dissipation

NumaChip

  • Scalable Server Building Block
  • ccNuma and Numa low latency shared memory interconnect
  • Virtualizes Everything, Including Memory and IO
  • >10x price/performance benefit over proprietary solutions
  • Seamless Scaling of Application Size and Performance - NO Porting Efforts

What is ccNUMA?
Cache-Coherent Non-Uniform Memory Access/Architecture (ccNUMA) is a computer memory design used in multiprocessors, where the memory is distributed across multiple nodes/processors where the processor cache is coherent to ensure data integrity and performance. ccNUMA architectures are the basis of symmetric multiprocessing (SMP) servers today.

With a new generation of ccNuma enabled interconnect, commodity x86 servers can be ‘clustered’ to form virtual SMPs..... at commodity prices!

Why ccNUMA?
As the HPC problem sizes becomes more complex and our scientist and engineers try to solve more complicated problems, PetaScale clusters will be the norm. Commodity x86 ccNUMA clusters are the only reasonable method to build and manage such a cluster cost effectively.

Complexity of Today’s HPC Clusters

.. all of our customers tell us managing a HPC cluster is complex!

  • Need to provision OS and software onto multiple nodes
  • Need to keep everything in sync
  • Node failures disrupts production workflows
  • We are scaling to Petascale.. how am I going to manage 10,000 instances of Linux OS?

ccNUMA - Best features of Beowulf clusters and SMP servers

  • SMP simplicity and management of the system - only ONE OS to manage
  • Compatibility with existing parallel and non-parallel applications including enterprise applications (no code changes! no porting of codes to run MPI!)
  • SMP features at Beowulf/HPCC commodity pricing!

What will the Top500 look like in the next decade?

ccNUMA systems will displace large SMPs and beowulf clusters - and we will see a repeat of the above trend over the next decade on the Top500.