論文

There is a critical problem of metadata performance scalability in high-performance computing when accessing many small files from millions of clients. This paper proposes a design of PPMDS, a scalable distributed metadata server, to improve metadata performance and scalability. To efficiently manage hierarchical namespace of file system across multiple servers, it manages inode entries using a parent inode number and an entry name as a key. To update multiple inode entries atomically, it uses a nonblocking distributed transaction based on an algorithm of dynamic software transactional memory. It improves the metadata performance by introducing server-side transaction processing, multi-reader, and a shared lock mode to reduce the number of remote procedure calls and to prevent unnecessary blocking. A prototype implementation achieves 62,000 operations per second when creating files in a single directory using 8 metadata servers, which is 2.58x performance improvement compared to a single metadata performance.

This paper proposes and examines the three in-memory shuffling methods designed to address problems in MapReduce shuffling caused by skewed data. Coupled Shuffle Architecture (CSA) employs a single pairwise all-to-all exchange to shuffle both blocks, units of shuffle transfer, and meta-blocks, which contain the metadata of corresponding blocks. Decoupled Shuffle Architecture (DSA) separates the shuffling of meta-blocks and blocks, and applies different all-to-all exchange algorithms to each shuffling process, attempting to mitigate the impact of stragglers in strongly skewed distributions. Decoupled Shuffle Architecture with Skew-Aware Meta-Shuffle (DSA w/ SMS) autonomously determines the proper placement of blocks based on the memory consumption of each worker process. This approach targets extremely skewed situations where some worker processes could exceed their node memory limitation. This study evaluates implementations of the three shuffling methods in our prototype in-memory MapReduce engine, which employs high performance interconnects such as InfiniBand and Intel Omni-Path. Our results suggest that DSA w/ SMS is the only viable solution for extremely skewed data distributions, but this solution is only valid on systems equipped with high performance interconnects. We also present a detailed investigation of the performance of CSA and DSA in various skew situations.

This paper proposes and examines the three in-memory shuffling methods designed to address problems in MapReduce shuffling caused by skewed data. Coupled Shuffle Architecture (CSA) employs a single pairwise all-to-all exchange to shuffle both blocks, units of shuffle transfer, and meta-blocks, which contain the metadata of corresponding blocks. Decoupled Shuffle Architecture (DSA) separates the shuffling of meta-blocks and blocks, and applies different all-to-all exchange algorithms to each shuffling process, attempting to mitigate the impact of stragglers in strongly skewed distributions. Decoupled Shuffle Architecture with Skew-Aware Meta-Shuffle (DSA w/ SMS) autonomously determines the proper placement of blocks based on the memory consumption of each worker process. This approach targets extremely skewed situations where some worker processes could exceed their node memory limitation. This study evaluates implementations of the three shuffling methods in our prototype in-memory MapReduce engine, which employs high performance interconnects such as InfiniBand and Intel Omni-Path. Our results suggest that DSA w/ SMS is the only viable solution for extremely skewed data distributions. We also present a detailed investigation of the performance of CSA and DSA in various skew situations.