Netlist Demonstrating its Innovative HyperCloud Memory Solution at INTEROP

Virtualization is driving new levels of efficiency and performance in the data center. For example it has changed the approach to disaster recovery: It’s no longer necessary to install and maintain additional server boxes for redundancy – “virtual machines” can be readily backed up to any physical server in any location, reducing the need for investment in additional physical servers for failover. This, in turn, is driving consolidation – instead of having two dozen physical servers in a data center for redundancy, an operator can deploy just a handful of high density, high performance machines, each running numerous virtual machines, and gain new efficiencies in computing and energy consumption. Having larger boxes with more ports suddenly makes sense, whereas before it didn’t.

But putting more virtual machines per server requires greater core memory per server. Providing ample memory for High Performance Computing in virtualized environments, for example, has been a challenge, as applications such as Finite Element Analysis (FEA), used primarily for engineering of complex machines such as planes and automobiles, requires high capacity levels of memory as well as sustained memory access speed. Typically in order to run simulations using FEA you’d need multi-node computing clusters and additional software licenses. Running these resource intensive applications requires greater core memory as it is – thus it is easy to see how running these applications in a virtualized environment can seriously tax the host server.

Irvine Calif.-based Netlist has developed a new approach to boosting server memory with its new HyperCloud Memory offering. The company claims this new technology facilitates memory capacities up to 288GB for Intel (News - Alert) Xeon CPUs and up to 384GB for AMD’s advanced multicore CPUs, with 1066-1333MT/s speeds. Basically it’s a new approach to overcoming the memory and speed limitations of today’s 64-bit processors.

This company claims this solves the limitations of JEDEC RDIMMs to support memory capacities over 96GB at 1066MT/s or beyond 192 GB at any speed in 2P servers.

Yesterday at Interop (News - Alert) New York I spoke briefly with Paul Duran, director of marketing for Netlist, Inc. who explained how the company’s new HyperCloud Memory solution works. I have to admit he got a little too technical for me, making it challenging for me to put this into laymen’s terms, but basically it is a memory module that: “delivers cloud scale server virtualizaton with maximum memory performance -- accelerates HPC analytic and simulation runtime performance -- and breaks traditional memory capacity and speed barriers, doubling system memory.”

Duran said with this new product data center operators increase the number of virtual machines per server by a magnitude of 3 to 1. It’s an innovative approach he said to increasing the amount of DRAM memory required to run virtual machines – thus it is another approach for increasing performance and reducing the number of physical servers needed (i.e. consolidation).

With its ability to literally triple the amount of full-speed memory for a 2P server, this product promises to reduce the number of systems needed to support virtualized machines by as much as 66 percent. This, in turn, enables data center operators to increase VM density, thus leading to data centers that are many times more efficient.

As Duran explained, HyperCloud Memory uses a patented design and proprietary ASICs to overcome the limitations of today’s CPUs. The company’s “virtual rank” technology enables memory capacity to be doubled through Rank Multiplication while maintaining speed performance through Load Reduction.

This can yield huge cost savings for data center operators: Not only can they run three times the number of VMs per server, reducing the overall number of physical machines needed and cutting power consumption, they will also get better performance out of those machines. This consolidation also reduces the number of software licenses that would normally need to be purchased.

The company is demonstrating the power of its HyperCloud memory technology this week at Interop by running 100 virtual machines on a single, fully loaded, 24-slot 2P server with 384GB of DRAM. Specifically it is using an HP DL385 G7 dual socket server with AMD's (News - Alert) Opteron 8-core CPUs and 24 memory slots. Fitted with 24 16GB, 2vRank HyperCloud DIMMs, the servers are running vSphere virtualization software from VMware with 100 virtual machines, Linux, and Microsoft (News - Alert)-based host software.

For more information, visit www.netlist.com.