Finding “affordable visualization” with enough scalable horsepower to solve HPC's most demanding problems is often impossible for scientists and engineers who need visualization capabilities to analyze large data sets. Working to improve the accessibility and affordability of visualization solutions, HP introduced (November 15, 2005) the HP Scalable Visualization Array (SVA), a high-end scalable visualization solution that completes the company's Unified Cluster Portfolio's integration of computation, data management and visualization in a single, integrated cluster environment.
“Visualization,” explains Steve Briggs, HPCD's SVA product marketing manager, “is a critical capability enhancing the productivity and performance of HPC environments. To be of significant value to the HPC customer, visualization must be sharable, scalable, accessible, and affordable — attributes that were missing from the market until now.”
HPCwire: What business problems are you solving with this product?
Briggs: Traditional large-scale visualization solutions are too expensive to maintain and upgrade and are built on proprietary technology. Small-scale visualization solutions are limited to single workstations that feature large memory but bounded rendering speeds. We believe high-performance computing customers – using visualization for oil and gas, scientific research, simulation, data mining applications – want Linux cluster capacity combined with the capability to run huge data sets at an affordable price. The SVA does that.
HPCwire: Hasn't visualization always been around for clusters? What's new about this?
Briggs: Clustered visualization is relatively new. In fact, it wasn't until the 1990s that WireGL addressed the rendering portions of the problem. And the seminal paper on compositing, “Parallel Volume Rendering Using Binary-Swap Image Composition” [Ma et al. 1994], was published just 11 years ago.
What's new is that HP's SVA is designed to do for high-performance visualization what clusters have done for supercomputing, which is make it affordable and accessible. Our solution distributes the rendering and provides for parallel compositing that eliminates bottlenecks that impede visualization and under-utilize the rendering engines. The use of industry-standard components drives affordability. And, as part of the Unified Cluster Portfolio, we complement the visualization technology with tools, applications and support to ensure successful production deployment.
Equally important to customers is the availability of applications that can take advantage of the visualization cluster technology. Applications such as Wolfram Research's gridMathematica, Infiscape's VRJuggler, CEI's EnSight, Visenso's COVISE, open source Visualization TookKit (VTK), open source ParaView and others, offer users real comfort in working with well known, trusted applications, while obtaining performance that, just a few months ago, was either impossible to achieve or extremely expensive.
HPCwire: What products make up the SVA?
Briggs: The SVA consists of a cluster of HP workstations running Linux, commercially available, industry standard graphics cards and network adaptors, and an integrated software system. Each HP SVA node in the cluster contains a high performance HP workstation configured as either as a render or display node. Each workstation has one or more PCI Express 16x graphics cards. System software includes XC System Software, Scalable Visualization Array Software for configuration and job management, and optional HP StorageWorks Scalable File Share (HP SFS) software for scalable storage, and optional HP Remote Graphics Software.
One of SVA's strongest attributes is flexibility. The SVA scales to support diverse visualization workloads including multi-user, multi-tasking, and multi-sessions. It supports various visualization styles, models, and display systems including single screens, caves or walls. By the way, the SVA technology can produce a vast, high resolution display wall of 100 million pixels and more. HP's SVA works in three basic modes, as a cluster of independent workstations, as a cluster of synchronized workstations, as a sort-last compositing cluster, or as a combination of all three. Since the system offers job and resource management capability, customers aren't forced to choose a rigid configuration and can dynamically change their capabilities as their requirements change.
HPCwire: How is this solution different from competitive products?
Briggs: First, there is the flexibility that we just talked about. Second, the HP SVA is affordable because it is developed on state-of-the-art industry standards and open source technologies. This makes it the only off-the-shelf, high-performance visualization solution on the market. Third, the HP SVA is a true Linux cluster and integrates visualization, computation, and data management to solve the toughest HPC challenges, including offering remote visualization and visualization collaboration.
HPCwire: HP has a number of high performance visualization solutions, such as the SV7. What's new and different about SVA?
Briggs: The SVA builds on HP's decades of graphics expertise. (Editor's Note: HP acquired Apollo Computer, one of the original graphics workstation vendors, in May 1989.) Visualization is not a case of one-solution-fits-all and HP is fortunate to have a broad portfolio of solutions. Our new graphics workstation, the xw9300 Workstation, is the first commercially available workstation to support two high-end 3D cards simultaneously. The HP Visualization Center, sv7, allows for accurate, real-time visualization of complete digital prototypes, permitting designers to visualize models with life-like 3D realism. So, while the sv7 has features suitable for CAD/CAM/workstation visualization, the SVA is suitable for a multi-user environment with graphic features suitable for scientific visualization, modeling and simulation, as well as geophysical exploration.
HPCwire: What are the scalability challenges with clusters and visualization? What's so different about SVA?
Briggs: From interconnects to pixel networks, these visualization clusters must not only scale but have the horsepower to scale quickly. The HP SVA scales up easily by simply increasing the number of nodes in the cluster. As an extension of standard Linux cluster, the SVA behaves like a cluster. The integrated clustering capability simplifies administration and improves distribution of resources to multiple users.
Scientists running visualization and computation applications generate huge datasets, requiring significant rendering power for visualizing that data. To handle those challenges, the HP SVA supports open source and commercial visualization software packages that drive high-resolution multi-tile displays and immersive environments, permit a mix of compute, render, and display nodes, and allows the use of computation steering to visualize while computing.
HPCwire: Is this only a solution for Linux clusters?
Briggs: At this time, yes. As Windows HPC becomes available, it'll make sense to support that, too.
HPCwire: How affordable is the SVA?
Briggs: The technology takes advantage of COTS components, open standards and open-source Linux – leveraging the tremendous advances made in readily available processors, graphics adaptors, interconnects, networks, clustering and middleware. The HP SVA costs about half of competitive products – and that includes installation. The architecture is modular, scalable and flexible, which not only gives significant technical benefits in solving grand-challenge problems but pragmatic benefits in terms of manageability, reliability, upgradeability and affordability.
HPCwire: What about performance?
Briggs: As you know, gaming is driving volumes of graphics cards. NVIDIA is using the physics of gaming to introduce improved floating point computations in the graphics cards. Graphics cards are doubling or tripling in performance every nine months making for vastly powerful performance in the HP SVA and guaranteeing improving performance in the future.
HPCwire: What's the downside? It can't be perfect.
Briggs: There are some rendering algorithms that are more suited to SMPs than clusters. HP, of course, offers a broad choice of SMPs as well as clusters.
HPCwire: Didn't you already announce this product a couple of years ago? What happened? Is this a different version?
Briggs: You are probably thinking of the Sepia project, and about the work of HP's Collaboration and Competency Network (HP CCN). HP CCN is an on-going forum to facilitate wide-ranging collaboration, innovation, discovery, and competency-sharing between HP and high performance technical computing customers and partners. One of our topic areas is scalable visualization. There are opportunities at a variety of levels for interested parties to participate. For more information, visit http://www.hp.com/techservers/hpccn/index.html.
HPCwire: HP really promoted Sepia. What does Sepia add to this product?
Briggs: Let me explain about Sepia. Sepia was a research program for the TriLabs (Los Alamos, Lawrence Livermore, and Sandia), the ASC (Advanced Simulation and Computing) VIEWS (Visual Interactive Environment for Weapons Simulation) program. It challenged us to develop technology based on industry standard components, which would solve the problems of visualization which, at that time, required massive proprietary SMP machines. As our research on Sepia progressed, so did the performance of graphics chips, graphics cards, processors, and networks. Consequently, HP's development focus shifted to compositing with other technologies and developing industry-standard APIs for advanced compositing functions. CPU/GPU compositing handles spatial tiling, depth compositing, and alpha blending, meeting many of our customers' needs.
Eliminating the dedicated Sepia hardware card resulted in significant cost savings for customers. That said, our research on Sepia resulted in software and algorithms which helps to increase the SVA performance.
HPCwire: What are the HPC trends in visualization? What can we count on?
Briggs: What is true for HPC is true for visualization – that is “if more is better, too much is just right.” You can count on increased data sets, increased computational capacity, and an increased need to visually interpret terabytes to petabytes of data. That's why it is important to boost the value of visualization through increasing real-time interactivity, scalability, accessibility, and affordability.