insideHPC » The 411

The 411: eXludus

John West — Thu, 08 Jul 2010 11:58:40 +0000

The 4-1-1 is an irregular insideHPC feature where we highlight some member of the HPC ecosystem with a quick look at who they are and what they do. If you’d like your company to be considered for a 4-1-1 feature, drop us an email.

eXludus develops software solutions that add resource allocation intelligence to Linux systems that dynamically optimize multi-core server throughput without need for re-writing applications. Here’s the 4-1-1.

Who: eXludus is a software development company that has spent the past 2+ years developing a solution that allows existing applications to make better use of rapidly increasing processor core counts. The founder and lead engineers spent most of their careers in high performance computing — vendor and user side — developing OS schedulers and cluster management tools, parallelizing applications, and running large supercomputer sites. Realizing that: a) multi-core processors and most existing (serial/lightly parallel) applications are not well matched, b) parallelizing applications is costly, and hard, and c) that more cores will place more strain on shared system resources, they have been applying their knowledge to developing an easy-to-deploy solution that allows multi-core processors to be used more effectively, thereby increasing overall system throughput.

What: MCOPt software transparently adds a dynamic application profiler and resource allocation layer to Linux systems. MCOPt increases system throughput by applying queuing theory techniques to ‘best fit’ actual resource needs to cores and memory. With MCOPt, users can safely process more concurrent tasks per system and take full advantage of multi-core power, as MCOPt limits shared-resource contentions and oversubscription problems which can lead to performance degradation. MCOPt is simply installed as a kernel module and requires no application changes. The company will soon be releasing an application profiler database extension to MCOPt that maintains historical runtime information that details actual resource consumption.

Why (you care): MCOPt provides an easily installed mechanism that can provide immediate throughput gains for many applications, and makes it simpler for users to extract more value from multi-core systems. If you have source access and in-house parallelism skills (neither a given), re-writing apps is a costly multi-year effort and, even then, the rate of parallelism will in most cases fall short of the rate of core count increase. An improved resource allocation layer that makes intelligent scheduling decisions to cores and memory based on actual application needs can help get more work completed in less time.

The 411: ScaleMP

John West — Wed, 01 Jul 2009 14:52:26 +0000

ScaleMP develops server aggregation virtualization solutions with a specific emphasis in high performance computing. Their flagship product is vSMP Foundation. Here is the 4-1-1.

Who: ScaleMP was founded in 2003 and has been focused on applying the principles of resource aggregation (prevalent in storage and networking) and commodity computing (leveraging industry standard servers and interconnects) to address high performance computing (HPC) challenges. In one of his previous roles, the founder led a large IT organization of hundreds of engineers and programmers and managed a broad range of complex IT projects in security, systems and network infrastructure, gaining significant experience in scalable computing and helped formulate the ideas behind ScaleMP.

What: vSMP Foundation is a software solution that aggregates multiple x86 systems into a single virtual x86 system (“reverse VMware”), delivering an industry-standard, high-end symmetric multiprocessor (SMP) computer. By using software to replace custom hardware and components, it offers a revolutionary computing paradigm for scalable computing, reducing overall end-user system cost and operational cost and complexity. vSMP Foundation aggregates up to 16 x86 systems to create a single system with 4 to 32 processors (128 cores) and up to 4 TB of shared RAM.

Why (you care): vSMP Foundation delivers a very versatile solution for high-performance and technical computing environments, combining the best elements of traditional SMP systems and clusters. Compared to traditional SMP systems, it provides a better performing, significantly lower cost alternative. It also offers for cluster customers a simplified clustering infrastructure with a single operating system and storage infrastructure resulting in a simplified cluster installation and management. Customers also have the option to select the best of breed programming model (OpenMP, MPI) for the task at hand.

When: vSMP Foundation has been shipping since 2006 and is currently in its 3rd generation, with satisfied customers across the world and supporting standard x86 platforms from all of the top server vendors.

Where: Learn more at www.ScaleMP.com,by email at info@ScaleMP.com, or by phone at 1-877-MAX-vSMP.

The 411: HPC Platform

John West — Tue, 13 May 2008 13:07:58 +0000

HPC Platform, headquartered in Paris (France), develops HPC and low-latency solutions for Market data processing and complex financial products valuations. Its HPCP appliances leverage the performance of FPGA-based hardware systems with the use of Domain Specific Language (DSL) compilers and optimized core libraries. The HPCP appliances have a very modular and scalable technical architecture that enables them to fit well the targeted applications (for instance market data processing + monitoring + valuation/CEP functionalities + risk analysis).

Who: HPC Platform was founded in 2007 and resulted from research and projects beginning in 2005. For its solutions, HPCP focused on a dedicated and global “Platform environment” comprising a modular hardware appliance, DSL compilers, optimized hardware cores and standard connectivity (Gigabit Ethernet, PCI-Express, etc.) The co-founders have combined experiences in Finance, IT, electronics, and software and some of them have worked on HPC solutions for the medical and industrial market since 1995. For the last 3 years, most of the work has been done with leading investment banks.

What: The HPC Platform solution, for ultra low-latency market data processing and complex financial products valuations, are true appliances with an optimized modular architecture, DSL compilers and software environments that aims to enable customers to effectively use the reconfigurable hardware technologies (FPGA).

Why (you care): HPCP delivers an outstanding performance over market data processing with a latency of 1.1 to 2.0 microseconds for fast encoded Feed-Handling/Filtering and order book processing. For complex financial products valuation acceleration (from simple Monte Carlo European options pricing to complex products like Altiplano), HPCP delivers an acceleration of several orders of magnitude compared to multi-core servers: for example the price of a basket option composed of 3 underlyings is computed in double precision within 715 microseconds for 1M simulation paths.

When: The HPC Platform solutions are already being evaluated in production environments in partnership with a few selected clients. Some results have already been published on www.hpcplatform.com.

Where: Learn more at www.hpcplatform.com or contact them directly at marc.battyani@hpcplatform.com.

The 411: PSSC Labs

John West — Mon, 07 Apr 2008 19:43:33 +0000

PSSC Labs (Professional Service, Super Computers) delivers high performance computing clusters, workstations, servers and storage systems. PSSC Labs solutions aim to allow customers to focus on their work and not managing their computers. Here’s the 4-1-1.

Who: Founded in 1989, PSSC Labs has shipped over 600 industry recognized PowerWulf Clusters. Each PowerWulf Cluster includes the Complete Beowulf Software Toolkit (CBeST) to facilitate cluster management, monitoring, and maintenance.

What: PSSC Labs delivers complete, turn key, high performance computing solutions including PowerWulf Clusters, PowerServe Servers, PowerStation Workstation and RaidStation Storage Solutions. All systems are custom configured to meet the needs, specifications and budget of the end users.

Why (you care): PSSC Labs understands the importance of the work conducted by their end users. PSSC Labs takes the necessary time and energy to properly configure and test all systems before shipping. This extra care and attention ensures that the end user receives a system that allows them to focus on their work and not managing a computer system.

Where: For more information please visit www.pssclabs.com

The 411: Terascala

John West — Thu, 13 Mar 2008 14:26:00 +0000

Terascala develops high throughput, high capacity storage solutions for performance-based computing. Its RTS 1000 Lustre-based storage appliance leverages the performance and robustness of the Lustre file system while masking some of Lustre’s complexity and associated costs. Here’s the 4-1-1.

Who: Terascala was founded in 2005, focused on applying the principals of commodity computing (leveraging existing solutions, commodity components, and open source software) to the storage challenges of high performance computing. The founders, and many on the team, worked together previously at Network Engines, where they developed the first 1U server, and at Ammasso, the first company to deliver a working RDMA Ethernet NIC.

What: The Terascala RTS 1000 Run Time Storage Appliance is a Lustre-based storage appliance offering high capacity and high throughput for performance-driven applications. It is a true appliance, with an optimized hardware and software environment that aims to be simple for customers to deploy and manage.

Why (you care): As the use of multi-core chips and high performance networks become standard in today’s high performance environments, the need to have a storage solution that can deliver I/O performance to “feed the beast” has become apparent. Alternative solutions typically are either custom, or customer built, and more expensive.

When: The RTS1000 is shipping now.

Where: Learn more at www.terascala.com or contact them directly at info@terascala.com or 508-588-1501.

The 411: Cluster Resources

Chris — Mon, 21 Jan 2008 15:52:49 +0000

Cluster Resources develops workload and resource management software including the Moab Cluster Suite, Moab Grid Suite, and Maui Scheduler. Here’s the 4-1-1.

Who: Incorporated in 2001, Cluster Resources, Inc. began initial development in the mid-1990s by the founders of the company under the name Supercluster Development Group. Before it became Cluster Resources, The Supercluster Development Group created the Maui scheduler and maintained & developed the open source Torque Resource Manager.

What: Moab Cluster Suite,® Moab Grid Suite,® and Moab Utility / Hosting Suite™, Moab Cluster Builder an integrated out-of-the-box cluster deployment and management solution based on SUSE Linux Enterprise Server 10. Cluster Resources also enables you to create dynamic hybrid environments through a policy set of dynamic triggers that provision resources on the fly.

Why (you care): Four of the top ten of the world’s most powerful clusters/super-computers, including the #1 world’s fastest supercomputer, license Moab to manage their workload (November 2007 Top500 list). The Department of Energy’s National Nuclear Security Administration’s Advanced Simulation and Computing Program standardized on Moab workload and resource management software across NNSA’s high-performance computing systems. This includes the Los Alamos National Labs Roadrunner system, Oak Ridge National Labs Cray XT3 Jaguar cluster and NERSC Franklin Cray XT4 system. Moab is the only meta scheduler that can effectively schedule and control jobs on multiple resource managers including Torque, PBSPro, LSF, LoadLeveler, SGE and Microsoft Job Scheduler out of the box. Through utilizing Moab’s policy sets resources (network, storage, licenses, operating systems, clusters, databases, cooling) can be dynamically changed on the fly to meet workload demands. Moab provides a single point of job submission and management over cluster and grid environments over multiple resource managers.

When: Moab Cluster Suite,® Moab Grid Suite,® and Moab Utility / Hosting Suite™are available now. Moab Cluster Builder based on SLES 10 is in beta and available for trial by emailing info@clusterresources.com there should be a public release by mid spring.

Where: You can find them on the web at www.clusterresources.com or by email and phone at info@clusterresources.com and +1 801-717-3700, +44 (1223) 437134 US headquarters are based in Provo, UT at the Novell campus, EMEA headquarters are based in Cambridge, UK at the Cambridge Science Park.

The 411: Interactive Supercomputing

John West — Wed, 19 Dec 2007 18:28:05 +0000

Interactive Supercomputing Corporation develops Star-P, a software platform that is delivering interactive parallel computing power to the desktop. Here’s the 4-1-1.

Who: Interactive Supercomputing (ISC) launched in 2004 to commercialize Star-P, an interactive parallel computing platform. With automatic parallelization and interactive execution of existing desktop simulation applications, Star-P merges two previously distinct environments — desktop computers and high performance Computers (parallel servers, clusters or supercomputers) –- into one. Based in Waltham, Mass., the privately held company markets Star-P for a range of biomedical, financial, and government laboratory research applications.

What: ISC develops Star-P, an interactive parallel computing platform that enables users to code algorithms and models on their desktops using familiar mathematical tools — such as MATLAB, Python and R — and run them instantly and interactively on parallel high performance computers (HPCs).

Why (you care): Star-P eliminates the need to re-program the applications in low level programming languages like C, C++, and/or Fortran and does not require the user to know the message passing interface(MPI) to run on parallel computers, which typically takes months to years to complete for large, complex and computationally intensive problems.

When: Star-P is available now.

Where: On the ISC Web site, www.interactivesupercomputing.com, you can find additional information about the company and product, as well as recent news and links to videos where users discuss the technology and its benefits.

The 411: NICE

John West — Wed, 12 Dec 2007 01:41:14 +0000

NICE has a big presence in the European HPC market, particularly on the industrial side. Here’s the 4-1-1.

Who:10 year old NICE is an Italian software company with offerings in HPC and grid management. Their software is aimed at company-wide management, optimization of computing resources, increased usability, reduced complexity for grid solutions. NICE headquarters are based in Asti, in the North of Italy, located in the medieval castle of Cortanze, a small pleasant village in the country-side, South East from Turin. Awesome…a castle.

What:NICE has one primary software products that can be customized to meet individual user needs. EnginFrame solves problems related to computing resource optimization and rationalization, streamlining the time and resources needed to deploy and configure a complete Grid solution in almost any heterogeneous environment. There is also a Grid Portal component that provides an efficient infrastructure to put Grid-empowered applications on your corporate Intranet/Internet. Application interfaces can be tailored to the specific users’ skills or access rights. Users can therefore access and control their computing and engineering resources via an intuitive, standards compliant Web interface, virtually from anywhere using a standard Web browser.

Why (you care): NICE delivers user-friendly, highly customizable access to Grid-enabled applications and infrastructures. NICE products tune up the Grid solution by increasing its usability and user-friendliness, without sacrificing flexibility and control for the most advanced computing scenarios.

When:EnginFrame and EnginFrame Grid Portal are available now. NICE products are deployed across a broad spectrum of companies in Europe, including Hess, British Gas, Statoil, Audi, Ferrari, Toyota, Alcatel, Ericsson, and many more.

Where:You can find them on the web at www.nice-italy.com and www.nice-software.com, or by email and phone at info@nice-software.com and +39 0141 901516.

The 411: Supermicro

John West — Thu, 01 Nov 2007 04:06:16 +0000

Super Micro Computer, Inc. is a motherboard manufacturer that’s expanded into systems and support infrastructure for servers. Here’s the 4-1-1.

Who: Established in 1993, Super Micro Computer, Inc. (NASDAQ: SMCI) emphasizes motherboards based on Intel’s x86 architecture, plus complete server systems utilizing Intel’s Itanium, Pentium, and Xeon microprocessors, as well as AMD’s Opteron processors.

What: Two Supermicro products are primarily of interest for high performance computing applications. First, the SuperBlade features 10 server blades, 4 switches, and 4 redundant power supplies in 7U enclosure. It supports both Intel and AMD Quad-Core/Dual-Core processors. Both Intel DP & AMD DP/MP processor-based server blades can be used in the same 7U enclosure. Second, the 1U Twin contains two dual-processor servers in a 1U chassis and supports Dual Intel Quad-Core/Dual-Core Xeon processors per node. The optional onboard InfiniBand connection is ideal for HPC clusters requiring high-speed low-latency communication between nodes.

Why (you care): With 960 processor cores, 24 Gigabit Ethernet, and 4X DDR InfiniBand, SuperBlade is a dense, energy efficient solution. The 1U Twin reduces total cost of ownership also by providing a high-efficiency power supply. The placement of two nodes in 1U also reduces the costs for chassis, power supply, cooling requirements, rack space, and IT room space.

When: Both the SuperBlade and 1U Twin have been shipping in volume. The SuperBlade with 14 blades per 7U enclosure will be available by the end of this year. 10Gb Ethernet switches for the SuperBlade will be available in Q1 2008.

Where: Super Micro’s OEM partners have had a number of wins in both the enterprise and HPTC markets, including Fermi Lab (130 nodes), India University (144 nodes), Jet Blue, and Jefferson Lab (300 nodes). You can find them on the web at www.supermicro.com.

The 411: Digipede Technologies

John West — Sat, 29 Sep 2007 02:52:28 +0000

Digipede Technologies is one of the companies making supercomputing on Windows a reality for enterprise customers with big requirements but without all the specialize knowledge and resources HPC has historically demanded. Here’s the 4-1-1.

Who: Digipede launched its flagship product at DEMO@15 in 2005. Digipede Technologies delivers grid computing software that enables enterprise applications to be distributed across a network of Windows desktops, servers, and cluster nodes. Digipede’s top vertical market is financial services, with applications ranging from risk management to trading analytics and pricing of complex assets; leading hedge funds and asset managers are among Digipede’s many repeat customers. Digipede also serves customers in government and defense, life sciences, entertainment and media, manufacturing, energy, and other markets. Competitors include Platform Computing, DataSynapse, and United Devices.

What: Built entirely on .NET, the Digipede Network works with Microsoft tools including Excel and Visual Studio. Digipede emphasizes the ease of adapting applications to its grid, providing a free Developer Edition that includes its acclaimed Digipede Framework SDK. By focusing exclusively on the Windows platform, Digipede provides a product offering far better integrated with the Microsoft technology stack than competing offerings.

Why (you care): Windows Compute Cluster Server 2003 builds a cluster in the sense that traditional HPTC users think of a cluster (MPI and all). But this doesn’t really help traditional Windows developers with existing business apps. The Digipede Network can put your CCE nodes at the disposal of a .NET developer without restructuring your application. By automatically deploying .NET assemblies (and related files), then distributing and executing .NET objects natively, the Digipede Network adds support for high-performance .NET applications to Windows Compute Cluster.

When: Digipede Network 2.0 is available now.

Where: www.digipede.net. You can also get in touch with the founder and CEO, John Powers, at john digipede net.

The 411: Woven Systems

John West — Wed, 29 Aug 2007 15:55:48 +0000

Networking startup Woven Systems has been getting increasing attention over the past several months in our community. HPCwire included them in their half-year retrospective on HPC, and their adaptive routing experiment with Sandia and Chelsio has also gotten a lot of attention.

I spent some time with them yesterday getting a product overview. I intend to write a longer summary of what I learned, but for now I just want to give you the 4-1-1.

Who: Woven Systems was founded in 2003 by networking industry veterans and has run on $15M in Series A venture funding (led by Goldman Sachs and Palomar Ventures) since 2005.

What: The company produces a 144-port 10 GbE Ethernet switch which they can use to create an Ethernet fabric that scales 4,000 edge ports while staying in Layer 2. This magic that makes it all work is Woven’s custom VScale ASIC. The ASIC also handles the heavy lifting for Woven’s other big advantage: adaptive routing in the network to avoid congestion. The company’s initial target markets are HPC (cluster interconnect and clustered file systems) and Internet Data Centers (data stream aggregation).

Why (you care): 10 GbE has not used in large clusters because 10 GbE switches are expensive, and building a cluster of any size would entail adding Layer 3 switches into the mix which introduce severe latency. The adaptive routing bit also offers a lot of promise in communication-intensive application over statically-routed solutions like IB (as shown in the Sandia results (text here, informative graph here. Woven claims there solution is 1/5 the power, cost, and rack space of existing solutions.

When: Their first product, the EFX 1000 Ethernet Switch, has been in trials at 5 sites and is transitioning to general availability now. As the company gears to move from development to production they’ve swapped out CEOs.

Where: www.wovensystems.com