| |
| |
Presentations
QsNetII Performance
By Quadrics
This presentation contains measured results for MPI and put/get communication including:
MPI Latency and Bandwidth
Batched RDMA Bandwidths
Put / Get Latencies
Barrier / Broadcast
Reduction / Gather
All-to-All / Bi-section bandwidth
Results are for AMD Opteron, Intel IA64, HP IA64 and Intel EM64T nodes.
|
The Roar of Thunder
by Robin Goldstone
Lawrence Livermore National Laboratory
presented at May 2004 Gelato Meeting
Lawrence Livermore National Laboratory (LLNL) has long been at the forefront of high-performance scientific computing. LLNL houses some of the world's largest supercomputers, with an aggregate peak capability of nearly 90TF. Over half of this capability currently comes in the form of parallel Linux clusters. In October 2003 LLNL solicited bids to build its largest Linux cluster to date: a 1024-node system with quad Itanium2 processors and the Quadrics Elan4 high-performance interconnect. The culmination of this effort was the achievement of a Linpack result of 19.94TF (nearly 87% of the 22.9TF peak) in April 2004. This talk will provide an overview of LLNL's Linux cluster strategy and then chronicle the birth of the Thunder Cluster, including the RFP process, build and integration, software development efforts, Linpack benchmarking, lessons learned, and future directions. |
Quadrics QsNetII: A network for Supercomputing Applications
David Addison, Jon Beecroft, David Hewson, Moray McLaren (Quadrics Ltd.), Fabrizio Petrini (LANL)
QsNetII is the next generation of the Quadrics interconnect. It consists of two ASICs: Elan4 and Elite4. The Elan4 communication processor forms the interface between a high performance multistage network and a processing node containing one or more CPUs. It has a 64 bit internal architecture and supports 64 bit virtual addresses. The Elan4 generates and accepts packets to and from the network. In addition, it provides local processing power to implement the high-level message passing protocols required in parallel processing. The network is constructed from Elite4 switch components, that are capable of switching eight bi-directional communications links. Each communications link carries data in both directions simultaneously at 1.3 GB/sec. The link bandwidth is shared between two virtual channels. The network supports broadcast transmission across selected ranges of nodes in addition to point-to-point connectivity between arbitrary nodes. |
Achieving order through CHAOS
The LLNL HPC Linux Cluster Experience
R.L. Braby, J.E. Garlick, and R.J. Goldstone
Early in LLNL' s Linux effort, a relationship with Quadrics was established to port device drivers for the Quadrics QsNet interconnect from Tru64 to Linux. This collaboration has been very beneficial over several years, and LC has exclusively used QsNet in its capability Linux clusters.
|
Evolutions of Interconnects for Supercomputing
by Quadrics
The requirements of supercomputing applications have driven the interconnect designer to deliver the best possible short message latency, and the most extreme compute communications ratios. With the latest interconnects delivering latencies of a few microseconds further improvements in latency are not possible without substantial changes in node design. Presentation given at the Trilab in April 2003. |
The Quadrics Network
High Performance Clustering Technology
Petrini, Feng, Hoisie, Coll, Frachtberg, LANL
The Quadrics network extends the native operating system in processing nodes with a network operating system and specialized hardware support in the network interface. In so doing, it integrates individual node's address spaces into a single, global, virtual-address space and provides network fault tolerance.
|
> Legal
| |
|