MIssion Critical Clusters
The slide deck white paper discusses the reliability pitfalls of status-quo networking technology, the penalties they impose on end-users and what better technology alternatives are there that industry and data center decision makers can adopt to make their cluster infrastructures as mission critical as possible.
Architectural Challenges of Cluster Networks
The slide deck analysis dissects the issues and opportunities that cluster network topologies carry, by directly comparing two popular network toplogies, i.e. Fat Tree topology, typically used with Ethernet and InfiniBand implementations and Torus topology, used for Native HyperShare cluster networks and supercomputer systems.
HyperShare Hardware and Software Brief
The paper provides an in depth description of the most important application of HyperShare, namely its use as a high speed interconnection network (often called a system area network) for computing clusters, data center servers and System-on-Chip (SoC) micro-server applications.
Time to Think Differently - "Good Enough" Clustering Technology is Out,
Distributed Resource Computing is In
This overview shows the path from the limitations and dangers of conventional data center cluster technology to the most infrastructure costs-, operating costs-, power usage-, modularity- and scalability-optimized cluster architecture - which we have termed Distributed Resource Computing - that HyperShare enables in full.
HyperShare over Ethernet - A Scalable, Commodity Standard
for Resource Sharing in Data Center
The overview presentation (Intl. HyperTransport Symposium, Mannheim, Germany) and related white paper by the Georgia Institute of Technology discuss the encapsulation of extended HyperTransport packets (HyperShare packets) into Ethernet protocol over standard Ethernet networks, thereby leveraging the popularity of Ethernet to enable sharing of cluster resources and (non-coherent) memory across the data center. More specifically, this paper describes key aspects of the HyperShare over Ethernet (HSoE) specification - in the document still called HToE - that is part of the HyperShare strategy. Best to view the presentation first.
Extending HyperTransport Protocol for Improved Scalability
The paper discusses why and in general terms how the HyerTransport protocol addressing capability can be extended to expand its reach from in-chassis CPU-to-CPU and CPU-to-I/O links to a powerful cluster network interconnect, with practically unlimited node addressing and cluster scaling capability. A much more specific and detailed version of this white paper can be accessed by HyperTransport Consortium members from the "Specifications" page.
Scalable Computing – Why and How
The paper explores the challenges of scalable, virtualized computing and how HyperTransport delivers system scalability and cross-system resource sharing capabilities that answer the industry’s enterprise and scientific computing requirements.
System Impact of Integrated Interconnects
This overview, presented at the HyperTransport Developers Conference illustrates part of the Consortium groundwork on global shared memory for the development of HyperShare networking technology. It touches on the evolution of cluster interconnect interfaces and elaborates on Partitioned Global Address Space Systems (PGAS).