|
High Performance Computing - Linux Clusters
Integrity Linux Systems has recently relocated. Our new facility was spec'd so we are able to power and cool as many as 512 processors in a single Cluster. We are the only privately held Linux systems provider in the state with this capability.
Integrity 160 node Myrinet Cluster.
(a gif image of a recently delivered Cluster that placed in the top 500.)
Integrity 16 node minicluster.
(a gif image of our mini-cluster turnkey system.)
Integrity Linux Systems Charter for HPC Clustering:
- To provide the most stable compute platforms, interconnect layer and network topology for HPC Clustering
- To provide the best tools and integrated system designs for your Software code developers
- To provide the best solution possible based on technical and financial considerations as outlined by our customers
- To provide superior technical support for our customers ensuring quickest deployment, highest uptime and resource availability to maximize productivity for parallel code developers
Integrity Linux provides COMPLETE TURNKEY HPC CLUSTER SOLUTIONS with unsurpassed reliability and manageability. Our customers include some of the most prestigious National Laboratories, Federal agencies and Commercial Fortune 100 businesses.
CLUSTER BRIEF: Linux HPC Clustering has quickly become the standard for parallel supercomputing. The Beowulf Cluster concept provides for the power of (up to) thousands of computer nodes to work in parallel to solve computational problems and render complex simulations. Complex problems in Science and Industry are addressed by utilizing high speed networking and high performance interconnects combined with relatively low cost systems achieve parallel computation with ever decreasing price for higher performance systems. Parallel programming languages and compiler tools are designed to use cost effective systems to solve complex problems. The price/ performance economy of scale between LINUX parallel computing topologies and traditional tightly coupled CPU/ Shared memory Super computing solutions is an order of magnitude of savings in most cases.
Once installed, the parallel code programmer has the opportunity to break down large complex problems into smaller sizes dramatically reducing computational time. MPI-CH, LA-MPI and OpenMPI parallel computing programming software environments are all supported under Linux HPC clustering. The addition of parallelizing compiler tools (such as Portland Group Compilers) has improved the ease of use of Parallel Computing Clusters. Many legacy parallel codes (example; MCNP, CTh) will run effectively in the realm of Linux Clustering.
High Performance Interconnects:
The Myrinet 2000 interface is a switched topology interconnect that is the most widely used in Linux Clustering. The PCI-X 64bit HBA card connects to the switch via industry standard fiber cable and provides a true 2GB/sec high performance interconnect with a sub 3 microsecond latency. A typical cluster on a single Myrinet switch is 256 nodes or 1024 CPU cores.The Myrinet CLOS256 14U chassis switch system components can easily scale to as many as 1024 nodes or 4096 CPU cores! HPC Software can include ROCKS, MPI-CH, OpenMPI, Portland Group Compilers, Intel Compilers, Low level hardware Command/Control and high level HPC Cluster monitoring tools. Myrinet 2000 is the most widely used HPC interconnect.
Myri-10G is a convergence that leverages 10-Gigabit Ethernet technology into the HPC world, and HPC techniques into the Ethernet world. The new Myri-10G solutions are 10-Gigabit Ethernet from Myricom, and more. Myri-10Gs Myrinet extensions open a broader universe of system solutions than Ethernet alone can provide. In Ethernet mode, Myri-10G network-interface cards (NICs) deliver maximal performance at minimal cost, and are fully compliant with Ethernet standards. Connect the NIC port to your 10-Gigabit Ethernet switch and you'll see up to 9.9 Gigabits/s UDP/IP data rate, or 9.6 Gigabits/s TCP/IP data rate. Connect the NIC to a 10G Myrinet switch and you'll unleash the capabilities of a network that supports the same software interfaces as Ethernet, plus low-latency, kernel-bypass APIs for Sockets, MPI, and DAPL. And, 10G Myrinet is interoperable with 10-Gigabit Ethernet for connections to IP networks (grids) and storage.
The Voltaire IB solutions provide a 10 gb/sec bandwidth with a sub 3 microsecond latency and scales to as many as 288 nodes without degradation of the interconnect. Integrity Linux Infiniband Clusters can speed the performance of the most complex Scientific problems by limiting latency times created by utilizing TCP over Infiniband eliminating TCP/IP packeting and bus sharing. Systems are delivered fully configured, burnt-in and tested. Software tools are configured with each Voltaire based IB cluster to provide at has now become a widely used HPC interconnect. Software includes MPI-CH, OpenMPI, Portland Group Compilers, Intel Compilers, Low level hardware Command,Control and HPC Cluster monitoring tools.
THE 'Glue':
The ROCKS Grid Cluster operating environment provides for an 'Appliance' approach to compute node management with a host of Open Source tools allowing for seamless and FAST HPC Cluster installation. The ROCKS approach treats the compute node as an appliance that requires NO traditional SYSTEM ADMINISTRATION. This approach reduces both the administration overhead and HPC Cluster setup/ debugging time. The Compute node system disk does not require it's Disk Image to be managed and can be brought down and re-built by invoking simple commands without bringing the cluster down . The head node is the only system that need be managed by a Cluster System Administrator. The Open Source ROCKS cluster software can run with either Red Hat Enterprise Server or a CENTOS free recompiled version of the Red Hat ES source.
The ROCKS Grid Computing model has been widely excepted as the premier Linux HPC Grid environment and has won 5 HPCwire readers and editors choice awards; Most significant HPC product 2005, Most Important Software Innovation 2005, Most Innovative Software 2004, Most significant HPC product 2004 and Most Important Software Innovation 2004.
The INTEGRITY LINUX Cluster Solution
INTEGRITY LINUX SYSTEMS melds the ROCKS Grid approach, HPC interconnects, other open source Clustering software tools and high density Storage, providing an integrated, rich Cluster solution. Our user interface provides for both low level hardware and high level Cluster wide command, control and system node monitoring applicable for any size workgroup or large scale HPC implementation.
Integrity Linux Systems welcomes the opportunity
to configure the Linux cluster technologies you need
to maximize your parallel computing power and reliability!
|
