At Delta HPC, we focus on supercomputing services for technical and scientific applications. Using a computer system to calculate quantum analysis requires a different approach and resources than processing data for business systems.
By choosing the services of Delta HPC, a science or engineering team can utilize twenty-first century CAVE visualization technology to test the data and analysis gained with supercomputing. This visualization and fast analysis can even be done simultaneously, so that the finished model can be tested virtually.
Our On-Demand business model permits your R&D teams to utilize the latest in supercomputing and visualization systems without having to invest millions of dollars in researching, purchasing and maintaining systems that you will not be continually using.
An application of how Delta HPC's focus of science and engineering provides a strong competitive resource can be seen in the development of a molecular model for a new drug. By combining Delta HPC's computational analysis resources for virtual screening or drug lead optimization with Delta HPC's CAVE facilities to develop a molecular model, the time for drug discovery can be accelerated.
Many of the current providers of high performance computing services are also resellers of supercomputer hardware. Others are providers of predominantly business services and their organizations are heavily tilted towards business expertise and concerns rather than accommodation for technological concerns and issues. It should come as no surprise that many of these same providers are shifting towards the use of commodity Linux cluster systems, where there is not much if any advantage over many of the systems that clients in the scientific community are currently using.
Delta HPC provides a world class highly secure computational facility for a diverse categories of firms that may be as large as multibillion dollar Global 1000 companies or that may be in the start-up phase. Your scientific and technical challenges do not conflict with another focus on business services and systems. Our objective is to provide a world class service. We treat every job as highly secure and confidential whether it is being run for the research team of a start-up or the R&D organization of a multinational.
Delta HPC's supercomputer center has been in operation since 2000.
Supercomputing or High Performance Computing refers to the use of very fast computational systems to solve complex scientific and engineering problems. Of course, thanks to Moore's Law, the envelope that defines a very fast system is constantly being pushed. Today's fastest supercomputers can now perform at rates approaching 300 TFLOPS (Tera Floating Point Operations per Second).
This will naturally vary with the amount of processing power and data storage requirements. For just the computer hardware system the cost may start at $250,000.
Other costs need to be added that include:
When all this has been calculated, the purchaser must consider the particular software, for which the system will be applied, which often costs more than the hardware to run it.
There are also the continued costs for IT personnel to operate and use the system and IT staff to provide 24x7 support. Supercomputers consume considerable power, so that there is a hefty annual energy cost to consider as well.
When all these costs are taken into consideration it is not unusual for even a very basic supercomputer system to have a total first year cost of $1 million and annual upkeep costs of $500,000.
The maintenance requirements for operating a supercomputer system include environmental control systems, and IT staff for system support and running applications.
Of the three main classes of general-purpose supercomputers, Delta HPC utilizes computers designed with architecture of Non-Uniform Memory Access (NUMA), which is currently the design of choice for the fastest supercomputers. NUMA computers utilize specially designed interconnects to have many processors and their memory communicate with each other. Processors and networking components are specifically designed for the machine.
Vector Processing architecture is another class, which allows the same arithmetical operation to be executed on a large amount of data simultaneously. Another and more common supercomputer architecture is that of Commodity Clusters. This architecture design utilizes a large number of commodity PCs, which are interconnected by high-bandwidth low-latency local area networks. This is the architecture of the Linux cluster system.
Delta HPC utilizes NUMA systems, which have a computer architecture that performs shared memory processing (SMP). SMP allows the memory to be combined so that large numbers of CPUs can access memory simultaneously. One CPU does not wait for another CPU to finish an operation as occurs in commodity Cluster architectures. Delta HPC's systems have been designed using Cache-coherent Non-uniform memory access (CC-NUMA). A system with CC-NUMA is in fact easier to program because of the aspect of fast access to a combined memory.
The NUMA architecture of Delta HPC's system means that the data in memory that constitute a job's tasks is viewed in its entirety by all processors at once. The connection to this data from the processor and between the processors is extremely fast relative to a Linux cluster. The result is that even with a supercomputer that has slower processors than those in a Linux cluster, the NUMA architecture advantage means that data intensive jobs will be performed much faster than in a Linux cluster system.
Specifically, the CC-NUMA systems at Delta HPC utilize SGI's Scalable Node (SN) architecture. In an SN system, processors, memory, and a bus- and memory-controller are coupled together into an entity known as a node, which is usually a single circuit board. These nodes are connected via a high-speed interconnect.
Because there is no internal bus whatsoever, access between processors, memory, and I/O devices is facilitated through a switched fabric of links and routers. This allows the combined memory of all the nodes to be accessed under a single OS image using standard shared-memory synchronization methods.
In a commodity cluster system such as a Linux cluster, the CPUs must wait for information to pass along LANs connecting the different CPUs before it can proceed to the next operation. Moreover, the CPUs do not have a complete view of the data in memory. The high speed 2 GHz CPU of a Linux cluster system must wait for the LAN and must stop to coordinate access to memory with the other CPUs.
In fact, an SN system is able to achieve a higher sustained vs. peak performance ratio than non-cache-coherent systems like conventional Clusters or Massively parallel computers, which require applications code to be written or re-written to do explicit message-passing communication between their nodes.
The distributed shared memory characteristics of the architecture mean SN systems are highly scalable. As CPU count increases, so does memory capacity, I/O capacity, and system bisection bandwidth.
As the number of nodes increase, which also means an increase in the number of CPUs that can be applied to a task, the more rapidly that useful work can be accomplished.
Delta HPC's supercomputer systems are built upon MIP processors, which are a family of RISC processors. RISC is short for Reduced Instruction Set Computer. Such a microprocessor design improves its speed and effectiveness so that only the most basic operations are necessary.
As with any computer system, there are limits to the performance features and capabilities of Delta HPC's supercomputers. Such constraints will include data size of the task, parallelizability of the job tasks, and the language codes and programs of the application.
Delta HPC has made sure in specifying its systems that such constraints do not limit the great majority of scientific supercomputing. For instance, a job would have to consist of many terabytes for it to be considered beyond the scope of Delta HPC's supercomputer systems. Delta HPC's systems were chosen so that it would be capable of running the great majority of scientific applications.
One constraint of supercomputers on speed of processing a problem is the amount of the task that is actually parallelizable. Amdahl's Law provides that speeding up a calculation through parallelization is subject to diminishing returns. Only that part of the problem that is parallelizable will be accelerated.
Amdahl's law is concerned with the speedup attainable from an improvement to a computation that affects a proportion P of that computation where the improvement has a speedup of S. Amdahl's law states that the overall speedup of applying the improvement will be:
For parallelization, Amdahl's law states that if F is the fraction of a calculation that is sequential, and (1 - F) is the fraction that can be parallelized, then the maximum speedup that can be achieved by using N processors is
The greater parallelization that is possible for a certain job and its tasks, the faster speed up advantage you will attain by using Delta HPC's supercomputers. A highly data intensive application that is highly parallelizable, such as a job utilizing ROCS or AMBER, will run several times more rapidly on Delta's NUMA architecture than on a client's Linux Cluster system.
On the other hand, not all jobs and tasks will be suited to Delta HPC's systems when Linux clusters are available to the client. Some jobs are more sequential and are therefore better performed on a Linux cluster system with high speed processors. For instance, a pattern recognition application such as BLAST is highly sequential. Consequently, many Linux cluster system will be able to run BLAST jobs at a faster rate than with Delta's supercomputer systems.
Delta HPC specifically chose SGI and HP systems over other system designs and brands. These systems have been highly respected and regarded by scientific and engineering professionals for several years and there are a great many scientific and engineering applications that run on these systems.
Your research teams may be using individual systems that utilize the latest in speed and design. Your R&D team may be locked into thinking that it can only do analysis on data sizes that the workstations permit. In fact your team members may be neglecting to do certain computational analysis because the members know that such data analysis will take far too long on the workstations or just aren't possible for the workstations to run. Delta HPC's on-demand model makes it possible and economical for R&D teams to analyze large data jobs and run them at accelerated paces so that the R&D process can be accelerated.
When you employ the services of Delta HPC, your job is given top priority and your data is given the highest standards of security and confidentiality. Your computational jobs and visualization tasks are handled by technical professionals with several years of industrial experience.
At an academic and research institution, your job will compete for time and expertise with the ongoing research of the institutions. The more open nature of these organizations also makes it more likely that your data and the investment that went into the research and development of your model are at higher risk.
Delta HPC's CAVE enables R&D teams to see their models and designs with an extremely high level of resolution and hence realism. Some of the best experts in the CAVE technology sector have commented that Delta HPC's system provides a sharpness and accuracy in imagery that is unsurpassed. This realism is possible because of the high quality stereo projector systems and the active operator tracking devices that are used in the CAVE. An image is obtained that just is not possible with a conventional workstation. In fact it is common for an R&D team that viewed the same image on a workstation to see new things and characteristics on the same image when viewed at Delta HPC's CAVE.
The CAVE is connected to a high performance SGI system that creates the digital image. The client also has the option of conducting on-the-fly supercomputing analysis on the object under consideration.
It is certainly true that in the last few years that the quality of images from visualization software running on workstations has improved considerably due to rapid advances in computer graphics, visualization software and the availability of powerful computer platforms.
With that said, the 3-D images obtained at a workstation are still small and flat.
Where such images require ample memory and processing power to view there is the additional length of time necessary to wait for various perspectives to be generated as modifications are conducted.
It is also more awkward to get a team of people to share an image and to have the individuals each collaborate on the image design.
In a CAVE setting, all team members can collaborate simultaneously as the image is modified.
The image or model can be viewed from all perspectives in a view as large as ten feet, so that even the minutest details can be scrutinized.
The team is able to view an image that can be viewed at a extremely high level of resolution and accuracy because of the utilization of a SGI computer that is dedicated to high resolution graphical image processing, the high quality viewing systems in the CAVE, and the innate design of the CAVE's 3D feedback stereo tracking system.
The images can be manipulated almost instantaneously by all team members because it is processed on a computer dedicated for graphical image processing. If necessary, data intensive analysis can be conducted on the model design, by linking the image directly to Delta HPC's supercomputers. Calculations can then be conducted on-the-fly for a research team to collaborate in real time.
Delta HPC's resources can be utilized for a wide spectrum of applications and sectors. In the past these have included aerospace/defense, biotechnology/biopharmaceuticals, computational fluid dynamics, medical devices, energy, environmental forecasting, seismology, and architectural/engineering.
If necessary Delta HPC can tap into its technical staff of professionals. This of course can be discussed on a case by case basis.
Delta HPC has been utilized by firms from hundreds to thousands of miles from our location in Cambridge, Massachusetts. Our facility can act as your own remote terminal. Data for supercomputing can simply be transferred to us remotely and securely using secure HTTP, secure FTP, or a VPN connection, depending upon the size of the data that must be downloaded and the complexity of the job management required. These modes have ensured the high speed delivery of client jobs and effective communication regarding computational analysis and the job results.
Your normal gateway and internet service will be adequate for many jobs. For larger jobs transfer of data can be arranged through higher speed connections such as a VPN. Some clients have preferred to send data storage devices and equipment such as a diskette, CD-ROM, DVD-ROM, tape or a hard drive.
Delta HPC provides a high security facility with multiple layers of physical, software and infrastructure protection for confidential data to prevent hostile and physical penetration.
Delta HPC's facilities are also ISO 9001:2000 certified so that all data processing are subject to a highly scrutinized set of procedures where the customers needs and satisfaction are the main objectives.
Running supercomputer systems, involves storage capacity on a petabyte scale. Consequently Delta HPC is able to provide hierarchical data storage services, dependent upon usage characteristics for your firm. Usage may include straightforward data storage, remote data warehousing, or data storage for several weeks for applications that must be run against very large volumes of data.