Building More Robust Systems
Steven J. Simske · 2013
In this chapter, the three forms of parallelism—by task, by component, and especially by meta-algorithm—are used to build more robust systems. It shows how a “sense” for knowing which parallel approaches are likely to improve system metrics—cost, performance, accuracy, and/or robustness, for example—can be learned through practice. Five different example topics are considered here: summarization; cloud systems; mobile systems; scheduling, and classification. A relatively deep consideration of summarization shows how meta-algorithmics can lead to a change in how ground truth is collected in the first place, and how the output of one meta-algorithmic system can be fed back into another system. The chapter revisits the cost model for processing information using a Constrained Substitution pattern, now accounting for the impact of a security breach. A very simple classification problem is exhaustingly elaborated in order to show some of the considerations that must be traded off when deploying meta-algorithmics. Controlled Vocabulary Terms robust control