The potential of a supercomputer choked by a network interface cannot offer adequate performance. This is the problem facing DARPA, the US agency working on the development of new technologies for military use. Jonathan Smith, Program Manager of the Information Innovation Office, talks about working with his team on a new technology called FastNIC.
DARPA: FastNIC for networks 100 times faster
Considering how the acronym NIC stands for Network Interface Controller, it is not difficult to imagine what the advantage offered by SuperNIC is: an increase up to 100 times the speed, so that the computational capabilities of the computers can be exploited to the maximum for operations that they range from mathematical simulations to scientific research, without having to slow them down when the machine is in the condition of having to exchange data with those that surround it, although within the same network.
The real bottleneck for the processor’s range is the network interface used to connect a machine to the external network, such as Ethernet, which severely limits data processing capabilities.
To give a concrete example, if the processor is able to chew on input and generate TB of information in a second in output, while in the same period of time the network interface cannot go beyond a few GB in transmission, it inevitably arises a bottleneck, a slowdown. Succeeding in the intent, DARPA would ensure a significant competitive advantage over other companies that manage supercomputers. However, the will is to make the initiative completely open source, so that a standard can eventually be created that everyone will benefit from.