At the International Supercomputer Conference in Dresden, Intel today officially launched two new technologies it says will take high-performance computing (HPC) to the next level of its development.
At a press conference at Silverstone last week, Intel previewed the two new technologies it says will greatly advance and accelerate the growth of HPC and server clusters – Intel Connects Cables and Intel Cluster Ready – to journalists from around the EMEA region.
Intel's Tom Willis says that as cluster systems have become dominant in HPC, so too has the problem of deploying and maintaining them, especially in terms of literally kilometres of copper cable.
"Clusters are basically racks of standard CPUs connected by cables," Willis says. "But as data rates get faster and faster, and the demand on cabling gets more extreme, then the problems that users are facing just get worse over time."
Intel Connects Cables enable Infiniband and 10GbE customers to achieve 20Gbps data rates and extend the reach between servers up to 100 metres, he says.
This means that customers can now design clusters based on their own unique business requirements without being inhibited by cable length, Willis adds.
Connects Cables also alleviates several deployment problems.
Traditional copper cables tend to be heavy and difficult to install, creating overloaded floors and racks and situations where airflow is blocked, negatively impacting cooling.
Willis says that, according to tests conducted by Intel, Connects Cables are 84% lighter than copper cables, 83% smaller, and have a 40% smaller bend radius – factors that hugely simplify the installation and expansion of clusters.
The compact nature of the cables and lack of optical tranceivers also reduces the overall costs involved in maintenance and installation.
"There are two main benefits from Connects Cables," Willis says. "Firstly, they allow customers to scale out their clusters a lot easier; and, secondly, they can dramatically reduce costs in terms of installation and maintenance.
"These two benefits alone will allow customers to build the larger clusters necessary for optimum high-performance computing (HPC)."
Intel Cluster Ready is a programme and technology that helps simplify the deployment, usage and management of clustered computer systems by providing a standardised and replicable way to build clusters and run off-the-shelf high-performance applications.
The programme includes a specification that sets minimum standards for software and hardware components enabling software developers to evaluate just once across many hardware platforms, vastly speeding up development and time-to-market readiness.
Cluster Ready also includes a software registration process for compatible HPC applications as well as a hardware certification process using Intel's Cluster Checker. This tool checks the cluster hardware and software components to ensure they interoperate correctly.
Cluster Checker also includes fault isolation, helping to improve early detection of cluster problems that can decrease productivity and increase support costs.
"In the past, HPC buyers faced increasing fragmentation and uncertaintly among the disparate mix of available cluster solutions," says Richard Dracott, GM Intel high-performance systems. "With Cluster Ready, organisations can purchase their clusters with the peace of mind of knowing that the equipment and applications are certified to work together.
"Instead of spending months defining and deploying a cluster, the customer experience is now more similar to that of buying an enterprise server," Dracott says.
Herbert Cornelius, director of Intel's Advanced Computing Centre, says the company is focusing on cluster technology because of its growing importance in HPC.
"Clusters are driving growth in HPC and currently account for 50% of this entire market," Cornelius says. "This is expected to grow to 85% or $11-billion by 2010. That's just on clusters in HPC.
"This is why we're interested in this area and, more particularly, why we're interested in clusters in the volume segement of HPC.
"Customers are saying they want systems that just work," Cornelius adds. "They want the best performance at the lowest cost. SMP machines have done a good job so far, but they are limited in scale and expensive.
"Cluster systems, on the other hand, are cost-effective and scale well, but they are difficult to deploy and use.
"Our response to this is technologies like Cluster Ready and Connects Cables," he says. "Technologies that will alleviate these problems and that are part of our cluster solutions platform vision for the future."