| OCR Text |
Show 117 4.2 Fault tolerance comparison The fault tolerance issues in applicative systems have been analyzed and means to achieve high reliability and availability have been suggested. Since a fault-tolerant measure of every system has an old fashioned competitor, the hardware triple modular redundancy method, it would be interesting to objectively compare these two approaches. Analytic, or qualitative comparison requires a model for the network and a mod.e l for the benchmark programs. A network model can be conveniently developed, as one example has been suggested in Chapter 2. The programming model seems difficult to formulate. Furthermore, the analytical fault modeling still needs to be defined and included in the overall analysis. Quantitative analysis based on simulation may be the easier comparison method. The primary measurement is the duration of program execution under both fault-free and faulty system conditions. The measurements produced by systems with fault-free assumption can reveal the overhead of a new approach. The data gathered from systems with embedded faulty components can demonstrate the efficiency of the new fault recovery method. 4.3 Distributed resilient structure The concept of functional checkpointing and the splice recovery mechanism can be adapted to other applications besides applicative systems. From a structural point of view, this fault recovery scheme establishes a distributed robust structure among a network of processors. A natural application of such a robust structure is in the 3rea of distributed databases. Keller and Lindstrom [38] tried the approach to implement a distributed database through functional programming, which can certainly use the resilient structure established in this dissertation. It is believed that the reverse ap-proach, which ports the robust distributed structure concept to database systems with imperative languages, is also a promising research· subject. |
