Facebook experiences 1,000 times more traffic inside its data centers than it sends to and receives from outside users. That traffic growth has prompted the social network to design a fabric networking architecture.
Amazon Web Services today announced Cluster Compute Instances (CCI). As Amazon’s own James Hamilton points out, the key piece of the CCI puzzle is the 10 GbE connection between cluster nodes. This differentiates CCI from previous HPC use in AWS, when customers couldn’t reliably run applications requiring low-latency interconnects. Furthermore, AWS claims “an 880 server sub-cluster achieved 41.82 TFlops on a LINPACK test run,” which means CCI outperforms the No. 162 system on the Top500 supercomputer list. Of course, the eventuality of this offering shouldn’t be surprising given Werner Vogels’ past experience with HPC and the high levels of interest in EC2 from the pharma/biotech community. One thing to watch, however, will be how HPC vendors, and RightScale, adjust their cloud-based cluster-management software to compensate for this new offering.
Yesterday I wrote about how Intel processors are dominant on the Interop show floor, proof that networking appliances are more inexpensive and powerful than ever before. We can add one more adjective to that description today — blindingly fast. In aggressive and dramatic fashion, Intel has entered the 10 gigabit Ethernet adapter market with a new line of dual-speed server adapters that use standard copper cabling and are priced under $1,000.
The new Intel 10 Gigabit AT Server Adapter will be available with the Dell PowerEdge product line, a popular commodity compute platform for server applications and networking appliances, and will operate at both 1 gigabit Ethernet and 10 gigabit Ethernet speeds. With this move Intel has effectively commoditized 10 gigabit Ethernet, which will undoubtedly drive the technology deep into enterprise IT networks.