| SGI Altix ICE is a high performance computing (HPC) platform designed for the most demanding scale-out workloads. This innovative new product line is based on a new blade architecture specifically designed by SGI to meet the unique needs of the HPC market. It provides breakthrough scalability, manageability, reliability and price/performance - without the compromises inherent in either traditional cluster or symmetric multi-processing (SMP) systems.
While small-node clusters can provide real value and flexibility at the small-to-medium scale, they have difficulty scaling for larger systems, where they falter on several critical issues: price/performance, data centre resource efficiency, reliability, and manageability. SMP systems, on the other hand, can address processing requirements in more challenging HPC environments, but are unable to take advantage of the cost economics of commodity components. SGI Altix ICE forges a new path between these two approaches, by providing a manageable, reliable, efficient, and highly scalable environment, at an unprecedented price/performance level.
The challenges of scaling today's computing environment
In recent years, the growth of HPC has been fuelled by the rapid adoption of clusters, driven primarily by their relatively low up-front cost and seeming ease
of scalability. As organisations attempt to scale their cluster environments to address ever more complex problems, however, they find themselves hitting a wall on both price/performance and scalability. While traditional small-node clusters prove useful and cost-effective for small and even medium-sized applications, users need to weigh the initial advantages of quick deployment and extensibility against the increasing complexity and cost that goes hand-in-hand with scaling these types of systems past a certain limit.
As clusters grow in size, the hardware and software costs for provisioning the nodes, deploying the applications, and managing, monitoring, and tuning the entire system to achieve anticipated performance goals can escalate dramatically - virtually eliminating the up-front cost savings and anticipated simplicity offered by commodity cluster solutions. The increasing complexity of HPC workloads has in fact created an ever widening gap in user productivity versus performance. SGI Altix ICE closes this gap, and was purpose-built to efficiently handle true HPC applications and large scale-out workloads. When evaluating a potential HPC solution, users should therefore carefully consider bottom-line effectiveness in four key areas:
- Price/performance
- Data centre constraints – power / cooling / space requirements
- Reliability
- Complexity and manageability
The benefits of clusters without the compromises
Designed for density, each SGI Altix ICE blade can accommodate up to eight Intel® Xeon® processor cores. An SGI Altix ICE can scale to 512 processor cores per rack, and 1000s of nodes per system. SGI Altix ICE also uses a high speed, low latency 4x DDR InfiniBand interconnect, integrated into a cable-free independent rack unit (IRU), which eliminates external switches.
In addition to its tightly integrated (yet highly scalable) hardware architecture, SGI Altix ICE ships with a complete, standards-based software solution stack, for maximum out-of-the-box functionality at a competitive price point. It therefore handles the critical issues inherent to cluster-based computing by offering unprecedented capabilities in price/performance, power / cooling / space efficiencies, reliability, simplified management and scalability.
Price/performance value
Its integrated blade approach enables SGI Altix ICE to focus all resources exclusively on delivering maximum performance for size and cost. With its unmatched performance density and low node cost, SGI Altix ICE offers an aggressive price/performance value proposition, especially in large-scale configurations. It also includes software enhancements to address a key performance issue often encountered in parallel systems: operating system synchronization. An SGI-engineered software mechanism synchronises operating system overhead to significantly improve performance on parallel workloads.
Power / cooling / space efficiencies
Built years ago for a very different computing environment, many data centres today suffer from limited power, cooling, and space capacity, meaning the environmental inefficiencies of traditional clusters are becoming prohibitive when scaling larger installations. SGI has long been a technology leader in solutions optimising power and cooling efficiency, and SGI Altix ICE utilises 90% efficiency redundant power supplies, combined with other high efficiency components, to minimise losses throughout the entire power architecture. This results in average electrical savings of 33% compared to more typical cluster implementations. (If data centre infrastructure efficiency is also considered, these savings can be doubled.)
SGI Altix ICE also employs a combination of high efficiency redundant blowers and optional water-cooled rear doors to deliver impressive cooling efficiency results. With the water-cooled option, SGI Altix ICE has minimal effect on ambient data centre temperature, since up to 95% of the rack heat is dissipated to chilled water. The water-cooled option significantly reduces cooling equipment power consumption - and also increases overall system reliability by mitigating the common problems of hot-aisle / cold-aisle recirculation and hot spots within the data centre.
In terms of space, SGI Altix ICE's breakthrough performance density results in up to 70% higher compute power density per floor tile (based on gigaflops per square foot) compared to other blade systems - while staying within data centre flooring constraints. Its performance density, combined with its highly efficient approaches to power and cooling, therefore ensure maximum utilisation of scarce data centre resources.
Reliability
Reliability is another area for concern with larger scale clusters. Compute nodes and other components inevitably malfunction over time, and cluster installations often lack sufficient redundancy to deal robustly with component failures. The networks tying the cluster nodes together can also suffer from reliability issues that grow exponentially as clusters scale. The complexity of scale-out cluster environments, with their multiplying points of failure, leads to further reliability problems. By comparison, SGI Altix ICE achieves a new standard for reliability through key features including:
- Diskless, hot-swappable blades
- Cable-free blade enclosures (IRUs), for reduced potential points of failure
- Redundant, hot-swappable system components
- High-efficiency power architecture, for reduced heat dissipation
- Optimal thermal design
- Fully buffered DIMMs, to reduce transient errors
- InfiniBand backplane, for high signal reliability
Integrated storage
SGI Altix ICE's 'diskless node' architecture removes storage from the compute blades - a design that decreases cost and power / cooling requirements while at the same time increasing overall system reliability. It also allows customers to choose the storage option that best fits their computing environment.
Simplified management and scalability
With its emphasis on component and software integration, SGI Altix ICE sets a new standard for simplicity in scale-out environments. On the hardware side, the clean design of the IRU, with its integrated blades, switches and interconnect, contrasts the ad hoc 'maze' of many cluster-based systems. This integrated approach continues with the complete, integrated software solution stack; hierarchical management architecture; industry-standard SUSE® Linux® Enterprise Server 10 (with planned support for Red Hat® Enterprise Linux®); factory integration and testing - for out-of-the-box deployment and immediate productivity.
FURTHER INFORMATION
www.sgi.com/altix/ice
|
