Did you know that it isn’t necessary for hackers to see your data in order to compromise it? The mere pattern that your computer uses to accesses its memory is enough to give away an alarming amount of information. To make matters worse, the risks tend to be higher on the Cloud as server space is shared with other users.
A hacker could potentially populate Cloud servers with code geared towards data spying, stealing crucial information as a result. That’s one of the main reasons a group of MIT researchers with the Department of Electrical Engineering and Computer Science has created a method to eliminate these types of cyberattacks. They’ve developed a way of disguising memory-access patterns, which the team is getting ready to implement into hardware.
Here’s how it works: instead of querying one memory address, a chip would call on a number of addresses to confuse potential hackers. This process, however, means a significant increase in the amount of data flowing in the memory.
Using a unique data structure
To address this issue, the team has relied on a structure dubbed “tree” to house memory data. It functions like a family tree in the sense that a node (a person) is only connected to one node above it (their parents), but may also be attached to nodes below it (their offspring). Addresses are assigned at random to a path through the tree, which confuses a potential data thief.
However, the chip still has to write data to the whole path after reading it from a single one. Why? So that a hacker can’t identify the node of interest.
“The root of the tree is a lot smaller than the bottom of [the] tree,” said Albert Kwon, an MIT electrical engineering and computer science graduate student. “So intuitively, you want to push down as far as you can toward the bottom, so that there’s no congestion at the top.”
Following the sequence of noted
When writing data, it is important that the chip follows the sequence of nodes in the path. A failure to do so would make it easier for a hacker to gain information. The researchers initially tried to accommodate this by sorting memory addresses depending on their locations in the tree.
“Sort is not easy to do in hardware,” said fellow graduate student Chris Fletcher. “So by the time you’ve sorted everything, you’ve taken a real performance hit.”
To solve this efficiency issue, the team summoned an extra memory circuit. It boasts storage slots capable of being mapped onto the sequence of nodes in any of the tree’s paths. A data block is stored at a relevant circuit spot when its location is determined.
Applications for the technology
Another key difference in the chip is the number of times it writes data. Instead of always writing data out after reading data in, the chip does so only every fifth time (it uses the other reads to rid itself of any decoy data). Therefore, when the chip writes data back out, it will typically have five outstanding data blocks to store on the final path.
When that’s not the case, the chip is equipped to handle logjams at the top thanks to the strategy of pushing data down the tree line.
If the system works without slowing data access, it may only be a matter of time before military and commercial applications become widespread, such as cryptography, finance, medical record keeping, secure communications and many others.