An Analysis of Linux Scalability to Many Cores
ABSTRACT
This paper analyzes the scalability of seven system applications (Exim, memcached, Apache, PostgreSQL, gmake, Psearchy, and MapReduce) running on Linux on a 48-core computer. Except for gmake, all applications trigger scalability bottlenecks inside a recent Linux kernel. Using mostly standard parallel programming techniques — this paper introduces one new technique, sloppy counters — these bottlenecks can be removed from the kernel or avoided by changing the applications slightly. Modifying the kernel required in total 3002 lines of code changes. A speculative conclusion from this analysis is that there is no scalability reason to give up on traditional operating system organizations just yet.
P.S. From MIT reseachers
Показаны сообщения с ярлыком Multicore. Показать все сообщения
Показаны сообщения с ярлыком Multicore. Показать все сообщения
вторник, 5 октября 2010 г.
четверг, 15 октября 2009 г.
JAOO: The Concurrency Revolution: The Hardware Story
Слайды с презентации Brian Goetz "The Concurrency Revolution: The Hardware Story" на JAOO Conference:
http://jaoo.dk/aarhus-2009/file?path=/jaoo-aarhus-2009/slides/BrianGoetz_TheConcurrencyRevolutionTheHardwareStory.pdf
Abstract: Everyone is saying we need a revolution in the way we approach concurrency; the availability of cheap multiprocessor systems has brought concurrency, which used to be a topic for experts only, into the foreground. How did we get here? Why are the performance-enhancing techniques that worked for twenty years no longer working, and how did this become such a problem that we have to change our programming models? In this talk, we'll dive under the hood and look at the hardware trends that have forced us to turn away from increasing sequential performance and towards multicore CPUs and chip multi-threading, and look at the implications for software design and performance analysis.
Target audience: Advanced engineers who are interested in performance management and tuning, trends in software design, or anyone who is interesting in understanding the magic under the hood.
http://jaoo.dk/aarhus-2009/file?path=/jaoo-aarhus-2009/slides/BrianGoetz_TheConcurrencyRevolutionTheHardwareStory.pdf
Abstract: Everyone is saying we need a revolution in the way we approach concurrency; the availability of cheap multiprocessor systems has brought concurrency, which used to be a topic for experts only, into the foreground. How did we get here? Why are the performance-enhancing techniques that worked for twenty years no longer working, and how did this become such a problem that we have to change our programming models? In this talk, we'll dive under the hood and look at the hardware trends that have forced us to turn away from increasing sequential performance and towards multicore CPUs and chip multi-threading, and look at the implications for software design and performance analysis.
Target audience: Advanced engineers who are interested in performance management and tuning, trends in software design, or anyone who is interesting in understanding the magic under the hood.
Ярлыки:
Concurrency,
Efficiency,
Multicore,
Performance
Подписаться на:
Сообщения (Atom)