An innovative connection for next-generation supercomputers from SLINGSHOT. Cray’s next-generation Shasta supercomputer represents a significant advance in the flexibility and capabilities of Cray supercomputers. The exascale class system will form the foundation of Cray’s combined simulation, analysis, AI, and data management architecture for the next decade and beyond.
And a key element behind these time-defining capabilities is Slingshot – Cray’s latest version of scalable interconnect.
Slingshot is made to act as a convenient network backbone, one that offers a variety of functions to comfortably distribute the Shasta system in the supercomputer and data center world. It is the most innovative, cost-optimized, and purposeful connection the supercomputer market has ever seen, designed to provide unprecedented scalability and performance for the most challenging work and computing and data management, helping users with closely related issues dealing with big data.
Cray implements a modified variant of Ethernet in Slingshot called HPC Ethernet. This is a proprietary protocol that addresses some of the shortcomings in the standard Ethernet protocol – namely, large headers and large packets, and the inability to scale to large HPC workloads. Slingshot lends some of the features found in a typical HPC network to Ethernet.
Slingshot uses Rosetta, Cray’s custom ASIC switch. Rosetta uses the standard physical Ethernet layer to support standard Ethernet protocols and to serve as a baseline for HPC Ethernet. On each connected device, Rosetta will negotiate the enhanced features of HPC Ethernet. The intention is to simply run standard Ethernet-connected devices over the Slingshot network, while the HPC-optimized traffic can operate reliably internally.