BlueDBM Provides Solutions to the Future of Cloud Networking

July 17, 2015 by Jennifer A. Diffley

The new system, invented by MIT researchers and presented at the International Symposium on Computer Architecture, makes FPGA flash storage servers as efficient as those using conventional RAM while saving energy usage on analytics of large datasets.

MIT researchers present a solution to FPGA flash storage cloud servers, making them faster and less power hungry.

Talk of cloud storage is omnipresent, with every company offering cloud solutions and scrambling to secure faster, secure, and more powerful servers. Traditional RAM cloud storage, however, is nowhere near as reliable as it should be: power outages, failures, and quirks are par for the course. 

Last month, researchers from the Massachussetts Institute of Technology unveiled a solution in the form of BlueDBM, an architecture to support and accelerate big data analytics. BlueDBM is  flash-based and has low latency with high-throughput inter-controller networks. The BlueDBM is organized into a rack composted of 20BlueDBM nodes, each of which consists of flash storage coupled with an FPGA. The nodes are connected to 8 other BlueDBM nodes through a high-speed serial link that can handle up to 10 gigs of bandwidth at 0.5us latency.

BlueDBM Storage Node

"...a rack-sized BlueDBM system is likely to be an order of magnitude cheaper and less power hungry than a cloud based system with enough DRAM to accommodate 10TB to 20TB of data.”

There is no way to handle complex data queries quickly unless it utilizes flash memory in a rack-sized cluster, and the paper released by the team behind BlueDBM details just how clunky and inefficient the current server systems are. BlueDBM is the solution to to this, and will enable companies like Google and Amazon to run big data without the bottlenecks and unreliability that plague current systems.