Interdisciplinary Series, Tuesday, April 15: 11 am – 12 pm
Location: CoRE Bldg, Room 431, Rutgers University, Busch Campus, Piscataway, NJ
Title: Distributed Storage Systems: Issues, Performance Analysis and Code Design
Speaker: Vinay Vaishampayan, DIMACS
Many service providers are offering data storage services to consumers for zero or very low cost. Thus cost minimization is vitally important to providers of storage services. One approach is to use low cost components along with erasure codes in order to enhance the reliability. Erasure codes have been well studied for their application to communication problems and bounds and constructions are well known in coding theory. However, the distributed nature of the application to storage systems leads to several new and interesting questions. I will consider some of these issues here.
Unlike a standard communication system, the failure mechanism in a storage system that uses erasure codes is quite subtle and requires careful analysis. I will address the problem of performance analysis, highlighting the important role played by the repair duration, i.e. the time to repair a failed disk. I will then address the problem of erasure code design with the added objective of minimizing the repair duration. After surveying previous work, a layered code construction based on combinatorial block designs will be presented followed by a discussion about the advantages of such an approach.