In the field of data sciences, a complex structure of lines and nodes that is representative of data is known as a graph. These graphs are used for a variety of purposes including study of social networks, performance of web pages, and analysing the structure of neurons in the brain. A graph can have billions of lines and nodes, thus, making it heavy in size and configuration. Multiple servers need to be used to process these heavy graphs, and the results obtained are successively used to draw inferences. However, researchers in the computer science and artificial intelligence department of MIT have designed a device that can easily process and decrypt graphs in a single computer only.
Optimization of New Device
The previous method of processing graphs was by running them through dynamic random access (DRAM) across a range of servers. The new device designed by the researchers uses flash lights used in smart phones to understand the graphs and process them. Although flash storage functions slower than DRAM, the researchers accelerated its performance by using flash chips and computation. Memory, computation time, and bandwidth also ges optimized with the use of the aforementioned arrangement in the new device.
Testing New Device
The device was tested on graphs with lines as high as 128 billion and up to 3.5 billion nodes. The conventional method of processing such graphs requires huge-size DRAM (up to 128 gigabytes) and the servers’ cost is exponentially high. The new device, on the other end, achieved the same performance with just 1 terabyte of flash and a gigabyte of DRAM.