2014 IEEE 33rd International Symposium on Reliable Distributed Systems Workshops
開催地
和文:
英文:
Nara
アブストラクト
The upcoming generation of computer hardwarebrought several new challenges for underling software. Thenumber of cores on a chip grows exponentially, allowing foran ever-increasing number of processes to execute in parallel.Therefore, how to efficiently utilize all of the concurrentprocessing ability provided by multi-core platform becomescritical for achieving good system performances. We proposea CARIC-DA framework which optimizes the performance ofDBMS on multicore platforms. Our middleware can improvethe performance of the private cache levels by dispatchingconcurrent database queries to run on different processorcores according to the data needs of each query. Extensiveexperiments from microbenchmark to TPC-C benchmarkhave confirmed the effectiveness of CARIC-DA in improvingthe cache performance and providing higher throughput forDBMSs on modern AMD multicore platform.However, the multicore platforms are becoming more andmore diverse and complex in recent years. Different manufacturesare making different efforts to boost their performanceby adopting different techniques. Therefore, it is important forsoftware to be well turned according to the specific hardwarefeatures in order to fully take advantage of the target multicoreplatforms. In this paper, we analyzed the efficiency of ourCARIC-DA framework on the modern multicore Intel platform.We compared the performance of different deploymentstrategies where we vary the mapping relationship betweendatabase processes and processor cores from Hyper-Threadingenable to disable configurations. We evaluated the impact ofCARIC-DA for different cache levels and for different querytypes on the Intel multicore platform.
各種検索
サポート
ヘルプ