Microsoft launches a "Big compute" feature for Windows Azure

At the 2012 Supercomputing Conference, Microsoft launched a set of "Big Computing" features for Windows http://www.aliyun.com/zixun/aggregation/13357.html ">azure" to support large computing applications. Examples of large computations include modeling complex engineering problems, understanding financial risks, studying diseases, simulating weather, or analyzing large datasets. Because cloud computing offers greater flexibility and economics, big computing is now increasingly turning to the cloud, which is what Windows Azure's "big calculation" is all about.

Large computing on Windows Azure provides a better infrastructure, including faster processors, more memory, and low latency high-bandwidth networks. This platform is ideal for high-performance computing (HPC) applications and other applications that require fast CPUs, large memory configurations, or high-speed network connections.

The entry-level configuration provides 8 cores of CPU and 60GB of RAM, and the other is 16-core CPUs and GB of RAM.

Both configurations include the following hardware systems:

Intel Sandybridge 2.6 GHz processor

DDR3 1600 MHz Memory

5 x 1 TB hard Drive

Two Receptacle 50x15

10 GigE storage and Internet connectivity
Connections between RDMA + InfiniBand (IB) Gbps nodes
Windows Azure's InfiniBand network supports computing remote Direct Memory access (RDMA) between nodes, which is currently the first technology used by Windows Azure in cloud computing. This technique allows applications to have no differentiation in memory access on multiple compute nodes, and the memory of multiple compute nodes is the same as for the application. The RDMA network provides excellent flexibility and cost-effectiveness in performance. For RDMA, Microsoft's test showed that a 4-byte packet was sent across machines with only 2.1 microseconds of latency, and network bandwidth was more than 3500 MB per second.

To showcase the high performance computing power on Windows Azure, Microsoft conducted a benchmark test of Linpack, gaining 151.3teraflops (167.7 peak) and 90.2% efficiency. This result ranked 165th in the nearest supercomputer 500 and ranked 27th in system efficiency. For users, this means getting faster results at a lower cost.