Having a slow connection is just too frustrating, but think how supercomputers feel.
All those hubs are doing all sorts of processing at lightning speed, but in the end, they serve on an old network interface to stay in sync.
DARPA does not like that. Hence DARPA desires to change it, especially by making a new network interface which is a hundred times faster.
What is the problem?
The problem is as DARPA considers it, processors and memory on a computer or server can work in a usual sense work at a speed of about 10^14 bits per second,
That is conveniently into the terabit region, and networking hardware like switches and fibre are competent of about the same.
DARPA’s Jonathan Smith explains that the actual bottleneck for processor throughput is the network interface.
Which is used to connect a machine to an external network, just like an Ethernet, hence critically limiting a processor’s data ingest ability.
That network interface usually takes the form of a card, making it a NIC and manages accepting data from the network and moving it on to the computer’s systems, or vice versa.
Sadly, its performance is typically more on the gigabit scale.
That delta within the NIC and the other components of the network implies a fundamental limit in how quickly information can be shared among different computing units,
For example, like the hundreds or thousands of servers and GPUs that execute supercomputers and data centres.
The quicker one unit can share its information with another, the faster they can go on to the next task.
Using the FastNIC program, DARPA wants to reinvent the network stack and enhance throughput by a factor of 100.
After all, if they can solve this problem, their supercomputers will be at an enormous advantage over others in the world.
In particular with those in China, which has competed with the U.S. in the high-performance computing arena for years.
However, it’s not going to be that easy.
The other central part will, usually, be redoing the software side to deal with the massive rise in the scale of the data the interface will have to handle.
Even a 2x or 4x shift would require systematic improvements; 100x will need pretty much a ground-up redo of the system.
FastNIC is barely just getting started, so let’s forget about it for now and see when DARPA can crack the code in a year or maybe three.
For more updates and the latest tech news, keep reading iTMunch.