ACM Transactions on Programming Languages and Systems (TOPLAS)
Characteristics of scalability and their impact on performance
Proceedings of the 2nd international workshop on Software and performance
Dynamic decentralized cache schemes for mimd parallel processors
ISCA '84 Proceedings of the 11th annual international symposium on Computer architecture
ACM SIGOPS Operating Systems Review
Dynamo: amazon's highly available key-value store
Proceedings of twenty-first ACM SIGOPS symposium on Operating systems principles
Critical sections: re-emerging scalability concerns for database storage engines
Proceedings of the 4th international workshop on Data management on new hardware
Shore-MT: a scalable storage manager for the multicore era
Proceedings of the 12th International Conference on Extending Database Technology: Advances in Database Technology
A new look at the roles of spinning and blocking
Proceedings of the Fifth International Workshop on Data Management on New Hardware
Proceedings of the 2nd ACM/SPEC International Conference on Performance engineering
| Hi-index | 0.00 |
Many-core systems differ from traditional multi-processor systems in performance characteristics and sometimes behave in unfamiliar ways for us. This can lead us facing performance troubles from unknown causes, requiring long time to fix them. To address those situations it is important for us to accumulate technology and know-how on performance of multi-core systems through a plenty of case studies. This paper describes a case study on unusual phenomena of a many-core based system in executing DBT-1 transactions with MySQL; where system throughput was decreased with the increase of the number of cores, an appearance of a scalability bottleneck. The root-cause analysis based on several measurements of the system has succeeded to reveal that the problem was caused from an interaction between two bottlenecks at the different layers, hardware and software, which is hardly found in traditional systems. The bottleneck at the hardware layer has been alleviated by introducing a newly developed lock mechanism that the analysis results suggested. Although the problem may be particular to the system, we, however, believe that the method and process to identify the root cause are of wide use for various scalability bottlenecks.