Big Data applications – once limited to a few exotic disciplines – are steadily becoming the dominant feature of modern computing. In industry after industry advanced instruments and sensor technology are generating massive datasets. Consider just one example, next generation DNA sequencing (NGS). Annual NGS capacity now exceeds 13 quadrillion base pairs (the As, Ts, Gs, and Cs that make up a DNA sequence). Each base pair represents roughly 100bytes of data (raw, analyzed, and interpreted). Turning the swelling sea of genomic data into useful biomedical information is a classic Big Data challenge, one of many, that didn’t exist a decade ago.
This mainstreaming of Big Data is an important transformational moment in computation. Datasets in the 10-to-20 Terabytes (TB) range are increasingly common. New and advanced algorithms for memory-intensive applications in Oil & Gas (e.g. seismic data processing), finance (real-time trading), social media (database), and science (simulation and data analysis), to name but a few, are hard or impossible to run efficiently on commodity clusters.
The challenge is that traditional cluster computing based on distributed memory – which was so successful in bringing down the cost of high performance computing (HPC) – struggles when forced to run applications where memory requirements exceed the capacity of a single node. Increased interconnect latencies, longer and more complicated software development, inefficient system utilization, and additional administrative overhead are all adverse factors. Conversely, traditional mainframes running shared memory architecture and a single instance of the OS have always coped well with Big Data Crunching jobs.
“Any application requiring a large memory footprint can benefit from a shared memory computing environment,” says William W. Thigpen, Chief, Engineering Branch, NASA Advanced Supercomputing (NAS) Division. “We first became interested in shared memory to simplify the programming paradigm. So much of what you must do to run on a traditional system is pack up the messages and the data and account for what happens if those messages don’t get there successfully and things like that – there is a lot of error processing that occurs.”
“If you truly take advantage of the shared memory architecture you can throw away a lot of the code you have to develop to run on a more traditional system. I think we are going to see a lot more people looking at this type of environment,” Thigpen says. Not only is development eased, but throughput and accuracy are also improved, the latter by allowing execution of more computationally demanding algorithms.
Until now, the biggest obstacle to wider use of shared memory computing has been the high cost of mainframes and high-end ‘super-servers’. Given the ongoing proliferation of Big Data applications, a more efficient and cost-effective approach to shared memory computing is needed. Now has developed a technology, NumaConnect, which turns a collection of standard servers with separate memories and I/O into a unified system that delivers the functionality of high-end enterprise servers and mainframes at a fraction of the cost.
- NumaConnect links commodity servers together to form a single unified system where all processors can coherently access and share all memory and I/O. The combined system runs a single instance of a standard operating system like Linux.
- Systems based on NumaConnect support all classes of applications using shared memory or message passing through all popular high level programming models. System size can be scaled to 4k nodes where each node can contain multiple processors. Memory size is limited only by the 48-bit physical address range provided by the Opteron processors resulting in a record-breaking total system main memory of 256 TBytes. (For details of Numascale technology see http://www.numascale.com/numa_pdfs/numaconnect-white-paper.pdf )
The result is an affordable, shared memory computing option to tackle data-intensive applications. NumaConnect-based systems running with entire data sets in memory are “orders of magnitude faster than clusters or systems based on any form of existing mass- storage devices and will enable data analysis and decision support applications to be applied in new and innovative ways,” says Einar Rustad, Numascale CTO.
The big differentiator for NumaConnect compared to other high-speed interconnect technologies is the shared memory and cache coherency mechanisms. These features allow programs to access any memory location and any memory mapped I/O device in a multiprocessor system with high degree of efficiency. It provides scalable systems with a unified programming model that stays the same from the small multi-core machines used in laptops and desktops to the largest imaginable single system image machines that may contain thousands of processors and tens to hundreds of terabytes of main memory.
Early adopters are already demonstrating performance gains and costs savings. A good example is Statoil, the global energy company based in Norway. Processing seismic data requires massive amounts of floating point operations and is normally performed on clusters. Broadly speaking, this kind of processing is done by programs developed for a message-passing paradigm (MPI). Not all algorithms are suited for the message passing paradigm and the amount of code required is huge and the development process and debugging task are complex.
Shorten Time To Solution
“We have used development funds to create a foundation for a simpler programming model. The goal is to reduce the time it takes to implement new mathematical models for the computer,” says Knut Sebastian Tungland Chief Engineer IT, Statoil. To address this issue, Statoil has set up a joint research project with Numascale who has developed technology to interconnect multiple computers to form a single system with cache coherent shared memory. Statoil was able to run a preferred application to analyze large seismic datasets on a NumaConnect-enabled system – something that wasn’t practical on a traditional cluster because of the application’s access pattern to memory. Not only did use of the more rigorous application produce more accurate results, but the NumaConnect-based system completed the task more quickly.
A second example is deployment of a large NumaConnect-based system at the University of Oslo. In this instance, the effort is being funded by the EU project PRACE (Partnership for Advanced Computing in Europe) and includes a 72-node cluster of IBM x3755s. Some of the main applications planned in Oslo include bioscience and computational chemistry. The overall goal is to broadly enable Big Data computing at the university.
“We focus on providing our users with flexible computing resources including capabilities for handling very large datasets like those found in applications for next generation sequencing for life sciences” says Dr. Ole W. Saastad, Senior Analyst and HPC expert at USIT, the University of Oslo’s central IT resource department. “Our new system with NumaConnect contains 1728 processor cores and 4.6TBytes of memory. The system can be used as one single system or partitioned in smaller systems where each partition runs one instance of the OS. With proper Numa-awareness, applications with high bandwidth requirements will be able to utilize the combined bandwidth of all the memory controllers and still be able to share data with low latency access through the coherent shared memory.”