Performance and cost comparison of mirroring and parity-based reliability schemes for video servers

In order to provide a reliable service, a video server must be able to reconstruct lost information during disk failure. We focus in this paper on reliability schemes that add redundancy within the server. We study two redundancy-based alternatives: mirroring-based schemes and parity-based
schemes. We analyze in this paper these two alternatives and propose their
adequate data layouts. We then compare them in terms of the server performance (throughput) and the costs of a single stream. Our results show that mirroring-based reliability is more cost effective than parity-based reliability.
We obtain the, at first glance, surprising result that a server that uses mirroring, which doubles the storage requirement, has lower per stream costs than a server that uses parity-based reliability.