Mehran Elyasi is a fourth-year PhD student at the University of Minnesota under supervision of Prof. Soheil Mohajer. He received a B.Sc. in Electrical Engineering and Mathematics from Isfahan University of Technology, Iran, in 2014.
Mehran is broadly interested in information theory and its applications in communication, distributed storage systems, and statistical machine learning.
An enormous amount of digital data is generated by Internet users on daily basis. Such a huge size of data is being processed and stored by data-driven applications. On the other hand, the number of users interested in accessing such data in key internet application, such as Facebook, is dramatically increasing. Distributed Storage Systems (DSS) are widely being used as the backbone of such large-scale storage systems, in order to provide reliability and data availability.
While storage units are individually unreliable and subject to temporal or permanent failures, the data must be protected, and made available for users’ access. This can be done by introducing redundancy in the data, which leads to an storage overhead. In addition, a considerable volume of network traffic is dedicated to the repair of failed storage nodes, as failures occur frequently in large-scale storage systems. While it is desired to simultaneously minimize the repair bandwidth and maximize the storage efficiency of the system, it is shown that there is a trade-off between them, and one can be optimized only at the cost of a loss in the other.
Mehran has designed a novel coding scheme, called Determinant Coding, for distributed storage systems. Our construction provides encoding/decoding algorithms for storage as well as an efficient mechanisms for the repair of failed storage units. These universally structured codes can operate in all the optimum points of the storage-bandwidth trade-off.