XPS: FULL: DSD: Collaborative Research: Moving the Abyss: Database Management on Future 1000-core Processors

点击翻译按钮获取中文摘要

There are fundamental problems to speeding up CPUs beyond what ispossible today. Because individual transistors are not getting anyfaster, manufacturers like Intel and AMD are no longer able to getmassive performance improvements just by increasing clock speeds(e.g., going from 2GHz to 10GHz). To overcome this, future CPUs willcontain hundreds to thousands of smaller computational cores on asingle chip which will all run at the speed similar to currentprocessors (e.g., 2GHz). This means that each single core will only beas powerful as current CPUs, but that the total aggregate power of allthe cores will be significantly more than what is possible today. Animportant problem with the advent of these new CPUs is that thedatabase systems that are used in all aspects of our society areill-suited for this change. Such database systems are used to storeand access data for a variety of applications, including on-linebusiness (e.g., Google, Facebook), scientific instruments (e.g.,astronomical telescopes), and medicine (e.g., MRI scanners). Thereason that they are not ready to handle these new "many-core CPUs" isbecause most of them use ideas that were designed in the 1970s and1980s when processors only had a single core. Thus, the purpose ofthis project is to develop both software and hardware technologiesthat will allow database systems to utilize the full computationalpower of future CPU architectures. The results of this project willenable organizations to deploy future applications on fewer machinesthat use less energy than what is currently used today.Computer architectures are moving towards an era dominated bymany-core machines with hundreds of cores on a single chip. Thisunprecedented level of on-chip parallelism introduces a new dimensionto scalability that current database management systems (DBMSs) werenot designed for. In particular, it becomes exceedingly difficult forthe DBMS to perform concurrency control, logging, and indexingefficiently. With hundreds of threads running in parallel, thecomplexity of coordinating competing reads and writes to datadiminishes the benefits of increased core counts. Thus, in thisproject we propose to develop a software-hardware co-design approachfor DBMSs in the many-core era. On the software side, rather thanattempting to remove scalability bottlenecks of existing DBMSarchitectures through incremental improvements, the PIs seek a bottom-upapproach where the architecture is designed to target many-coresystems from inception. On the hardware side, instead of simply addingmore cores to a single chip, the PIs will design new hardware componentsthat can unburden the software system from computationally criticaltasks.For further information see project web site at:http://db.cs.cmu.edu/projects/1000cores/