- 1 VMware Takes the Wraps Off vRealize Automation and vRealize Business
- 2 Microsoft Previews Hyper-V Containers for Windows Server 2016
- 3 Mirantis Led FUEL Project Gets Installed Under OpenStack Big Tent
- 4 Red Hat Enterprise Linux 7.2 Adds Security, DR Features
- 5 Docker Reaches Across Universes at Dockercon EU
Supercomputing With Microsoft
Microsoft's plan to use the Internet to provide software services may be garnering the lion's share of attention, but the software giant is pretty keen on making headway in a sector where it is still a relative unknown: high-performance computing (HPC). The software giant's new Windows Compute Cluster Server 2003 is aimed at serving departments and workgroups.
On Tuesday, the company unveiled Windows Compute Cluster Server 2003, a new piece of cluster software designed to bring powerful tools to customers that require advanced computing but don't have deep coffers.
The second beta of Windows Compute Cluster Server 2003 will include CDs to install Windows Server 2003 Compute Cluster Edition and Microsoft, both of which are designed to run on Windows Server 2003 SP1 x64 Edition, said John Borozan, group product manager for Microsoft's server division.
While most supercomputing solutions are pieced together and a mish-mash of different products and protocols, Microsoft aims to provide a complete Windows platform for HPC to help customers get up and running more quickly.
To wit, the software will support Ethernet and Infiniband protocols and is integrated with the following Microsoft technologies: cluster setup and administration, compute node management based on images, security based on Microsoft's Active Directory, and job scheduling and resource management.
Microsoft expects to target departmental and workgroup clusters in the $50,000 to $250,000 range for manufacturing, geosciences, life sciences, oil and gas, and financial services sectors.
Borozan said Microsoft's entry into the market is predicated on some rising trends, namely the falling cost of HPC and its increase in enterprise deployments.
The executive said the cost of running supercomputers has dropped significantly: In 1991, $40,000,000 would get a customer 10 gigaflops of compute power. Today, with so many machines performing in the teraflop range, the cost of 10 gigaflops is roughly $4,000.
HPC is dipping into mainstream commercial deployments. Market research from IDC found that HPC deployments grew 70 percent in 2004, with most of the growth coming in departmental or workgroup clusters that cost less than $50,000.
It also doesn't hurt that x86 server clusters, the area Microsoft hopes to serve, are growing.
"We think the opportunity is ripe for us to enter the market," Borozan said. "We believe the next big revolution in both science and industry is going to be data driven. There are mountains of data being produced by computational models or sensors, but until now the ability to do more with that data and crunch it in a fashion that is used in HPC has been out of reach."
Despite targeting a specific, lower end of the market, Microsoft's competition will be steep, with IBM, HP, Dell, and Sun waiting in the wings.
But Microsoft has already garnered support from 19 ISVs that have ported their applications to Compute Cluster 2003.
Moreover, Microsoft has pumped multiyear, multimillion dollar investments into joint projects at 10 HPC institutes worldwide. The groups, including the Cornell Theory Center and University of Virginia will do advanced research using Compute Cluster 2003.
Separately, Cisco Systems this week unveiled new HPC solutions. The new Cisco SFS 7012 and Cisco SFS 7024 switches use InfiniBand technology to provide a unified fabric for connecting servers together into grids of compute resources.
Working with the SFS advanced Ethernet and Fibre Channel gateway, the switches connect server grids with shared LAN and SAN resources connected through Cisco Catalyst switches or Cisco MDS 9000 storage networking switches.
Cisco also has new HPC software on tap in the form of the SFS Subnet Management software, which scales beyond 4000 server nodes and recalls a multi-thousand node InfiniBand cluster fabric in less than a minute.
This article was originally published on internetnews.com.