Today Symmetric Computing announced a new line of low-cost, aggregated shared memory systems. Described as a “Departmental Supercomputer,” Symmetric’s Duet is a 64-core, large shared-memory SMP server with up to 1 Tbyte of memory and 3 Teraflops of power.
The large system memory and programming simplicity of SMP supercomputers are critical. Researchers, scientists and engineers want to focus on their specialty – what they do best – and not have to become experts in the intricacies of cluster computing. To programmers, our Duet Departmental SuperComputer looks just like a single huge-memory Linux box. Programmers can use standard threading packages to access all 64 CPUs and up to 1 TeraBytes of memory. Scientists and engineers need no longer worry about message passing interface programming, which is what supercomputing clusters and other limited memory systems demand. There’s no need to build complex file-access program components; they can just read their big datasets into memory and access them as an array. Now with Symmetric Computing, engineers and researchers can afford the SMP supercomputing resource they really want, in-house, and stop wasting their time with overly complex programs.”
To create a single-system image from aggregated machines, the Duet pairs its patent-pending Distributed Symmetric Multi-Processing (DSMP) software with OpenSuSE Linux and off-the-shelf server components including two (2) 1U SuperMicro A+ 1042G-TF blade servers with eight (8) 2.6 GHz AMD Opteron 6140 processors (64 cores total) and sixty-four(64) 8-GB or 16-GB DDR3 DIMMs (512-GB or 1-TB RAM total). As a system interconnect, the Duet uses 40 Gbps Mellanox InfiniBand PCIe host bus adapters.
According to the company, DSMP technology is inherently faster than hypervisor-based solutions because it operates at the operating system level. With DSMP, all of Duet’s shared memory is at most 2 microseconds from all 64 cores.