The Service Rate Problem in Distributed Storage and Non-systematic Recovery Systems

Talk by Andrea Di Giusto

Date: 11.03.26  Time: 16.30 - 17.30  Room: Y27H28

Data centres are one of the backbones of modern-day digital infrastructure, with the amount of processed and stored data growing year by year. In this context, coding techniques are employed to ensure the availability and reliability of the stored data.  One question that has recently emerged, is how to optimally store a certain number of data objects on a certain number of servers in a way that allows the system to serve certain amounts of requests for the objects. This problem is naturally phrased as a coding question, as it concerns the design of a redundancy strategy for the storage system. In this talk, we will introduce and formalize the service problem and the Service Rate Region as a measure of system flexibility. Connection to other well-established research topics will be outlined, such as PIR/batch codes and Majority Logic Decoding. The difference between systematic and non-systematic systems will be considered, and we will show how the latter can give a better performance when designed carefully.