Design and Complexity of Distributed Algorithms

点击翻译按钮获取中文摘要

Distributed computing systems ranging from simple desktop PCs or even cell phones with multi-core CPUs to high-performance computer clusters and systems in the cloud have to deal with problems that arise from concurrency, asynchrony, and failures. In these settings, the design of time and space efficient algorithms, proof of their correctness, and analysis of their complexity are challenging tasks. Algorithms and data structures for fundamental tasks in distributed systems are critical for the future development of efficient and fault-tolerant software. Lower bounds and impossibility results can guide researchers and software developers in their efforts to build efficient software systems, and can help hardware designers to decide what hardware primitives are needed. In my research program I will devise new efficient algorithms and data structures for shared memory systems. I will try to find solutions for fundamental data structures such as concurrent dictionaries, and new implementations of strong synchronization primitives, e.g., multi-word compare-and-swap and snapshot objects. I will complement new algorithms with proofs of lower bounds in order to determine the time and space complexity of these problems. ******I will also investigate new algorithm design and analysis methods. The classical approach to allow no room for error may limit algorithm design, and can lead to inefficiencies. I will study concurrent Monte Carlo algorithms that may produce errors or faults in very rare cases, but perform better than their error-free counterparts. Related to that will be my study of the recently defined allocate-on-access space requirements of shared memory algorithms: Many fundamental problems are known to require a large amount of (memory or disk) space in the worst-case. But these worst-case scenarios may be very improbable. The significance of the worst-case space complexity is justified by an assumption that any space that is used in some execution must be allocated for all executions. This assumption is not consistent with the storage allocation mechanisms of actual practical systems. The allocate-on-access measure is a more realistic approach to measuring actual space usage on modern storage systems without being overly pessimistic.*******Results of this research program will yield a better understanding of the system requirements of synchronization problems. Algorithms resulting from the research can be used as building blocks in software development, and lower bounds can help software and hardware developers making design decisions. Research on novel Monte Carlo algorithms and the allocate-on-access space complexity will directly and indirectly yield new solutions to synchronization problems, and improve the performance of systems relying on such algorithms.