Quadrics

From Wikicompany

Quadrics Ltd. designs and manufactures high performance supercomputer interconnect technology.

The company produces hardware and software for clustering commodity CPU's into massively parallel systems. Quadrics main product QsNet, is based on internally developed ASICs, that links numerous Linux servers into a single cluster supercomputer.

Quadrics has its HQ in the UK and local presence in the US, China and Italy.

Quadrics is fully owned by Alenia Aeronautica, which is part of the Finmeccanica Group.

Products

Hardware

• QsNet I - Quadrics interconnect based around the elan3/elite3 ASICs (350MBs @ 5us MPI latency)
• QsNetII - Quadrics interconnect based around the elan4/elite4 ASICs (912MB/s @ 1.3us MPI latency)
• QsNet II E-Series- a range of small-medium configurations (8-128-way) at USD 1,700 price/per port

General Characteristics of QsNetII

• Low latency and high bandwidth.
• Scalability to thousands of nodes.
• 64-bit internal architecture.
• Reliable transmission protocol.
• Commodity interface.

Software

• Quadrics RMS - Cluster resource manager software package
• Quadrics Linux Software - Core components of the QsNet software release for Linux under the GNU Public License

Quadrics New Software Release

• includes support for Rocks v3.3.0.
• new version of RMS, RMS 3.x, the Quadrics cluster management toolkit
• support for RedHat Enterprise Linux 4 and Fedora Core 3
• tested on the standard Red Hat(TM) Advanced Server and SuSE(TM)
• significantly improves the performance of intra-node MPI with single-copy process to process memory copies without the need for shared memory FIFOs.

Quadrics kernel patches are available when user's production use require added functionalities such as the ability to communicate from pageable virtual memory without locking down memory used in inter-process communication; however the most recent releases of the Quadrics device driver modules provide support for use of the QsNetII (Elan4) hardware with standard, unpatched Linux kernels.

Performance Overview

The interconnect is really the supercomputing bone of a cluster. As more and faster computers are added, networking becomes more complicated. The effectiveness and total overall performance of the cluster supercomputer is related to how quickly the computers can communicate with each other, even when there are thousands of them.

• Latency: 1.3usec on AMD Opteron.
• Bandwidth: 912 Mbytes/s on Intel Xeon EM64T.

QsNetII performance including Bandwidth in Multi-Rails systems, Batched RDMA, Performance on Collectives (Barrier, Broadcast, All Reduce, Gather and All-to-All)are available in http://www.quadrics.com

Performance alone, however, is only part of the picture. Quadrics' unique solution lies in combining the performance of QsNet with a management software to ensure that clusters spanning thousands of processors can be used, administered and run effectively. Cluster are often cheaper than traditional supercomputers and also offer another advantage: with traditional supercomputers when customers require more computational power they need to buy a new supercomputer, with a cluster they simply network more computers onto it.

History

Quadrics formed in 1996 as a joint venture between Alenia Spazio and the technical team from Meiko Scientific.

Their first design was the Elan2 network ASIC, intended for use with the UltraSPARC CPU, attached to it using the UPA syastem bus. Plans to introduce the Elan2 were later dropped, and a new Elan3 hosted on PCI introduced instead. By the time of its release Elan3 had been re-aimed at the DEC Alpha/PCI market instead, after Quadrics had formed a relationship with DEC.

The combination of Quadrics and Alpha EV6 CPUs proved very successful, and Digital/Compaq (now HP) rapidly became one of the world's largest suppliers of supercomputers. This culminated with the building the largest machine in the USA, the 20TFLOP ASCI Q, installed at Los Alamos National Laboratory in 2002/2003. This machine consists of 2048 4*1.25 GHz CPU EV67 Alphaserver SC ES45 nodes, and two rails of the Quadrics QsNet network.

Quadrics also had success in selling Linux based systems, with the fastest Linux cluster in the world called Thunder installed at Lawrence Livermore National Laboratory in 2003/2004, Thunder consists of 1024 Intel Tiger Quad Itanium® II Processor servers to deliver 19.94 teraflops on parallel Linpack. Peak performance of the system is 22.9 teraflops. The level of efficiency is 87%.

In 2004 Quadrics was selected by Bull for what will be the fastest supercomputer in Europe: Tera10 at the French CEA: 544 Bull NovaScale 6160 computing nodes, each including eight Itanium® 2 processors. The global configuration will feature 8,704 processors with 27 terabytes of core memory. Each of these computing nodes will contain multiple Quadrics QsNetII (Elan4) network adapters to deliver over 60 teraflops (sixty thousands billions of operations per second).

More recently, Quadrics has been selected by HP for the upgrade of Sharcnet, the Canadian Cluster of Clusters, with four new high-performance computing clusters that will increase the network's capacity from 1,000 to 6,000 processors. QsNetII will be used for one capacity and one capability cluster.

In August 2005 Quadrics and STMicroelectronics signed a development agreement. The cooperation will cover the design of a future generations of Quadrics high speed multi gigabit interconnect, and the exploitation of the products in a range of high volume applications.

External links

• LLNL Thunder page
• Top500 website