New data loss model for RAID system

Contact: Luoming ZHANG
Room: BC 158
Tel: 021-693-13-13
Email: luoming [dot] zhang [at] epfl [dot] ch

Description: MTTDL is one of the most frequently used model in the storage reliability community. It's calculated by the Markov-chain model. In order to get a nice mathematical description,it's based on simplistic assumptions about the the physical system. So it has several drawbacks even though it's widely used as a gauge for storage system reliability. For example, the MTTDL model assumes all the devices in RAID to be new device after every reconstruction. Actually, this is not true which lead to some results that archive system has life time of 1400 years. The project is aimed at building up a new more reasonable model for RAID system reliability considering the continuous disk failure, the sector latent error and so on.

Requirement: Probability, especially good at Markov chains. Better with basic understanding of RAID system