Spark的网络传输是基于Netty的,而这个又依赖于 java.nio或者Linux epoll。虽然Android ART 内置就支持java.nio,但是我们需要重写Netty,使得我们可以在iOS上使用kqueue。除此之外,目前社区还不清楚是否低级别的网络原语(比如zero-copy)是否可以在JavaScript中使用,我们需要更密切地与苹果和谷歌合作,以改善JavaScript对移动网络的更好支持。
During transformation, images are displayed along with possible tags and users choose the correct tag for each image. To make sure the transformation is fault tolerant, we will randomize the ordering and retry 10 times for each record. Yes, RDDs are always deterministic. And of course, we support tagging with multi-labels, even on a wearable device:
In conclusion, this design doc proposes architectural changes to Spark to allow efficient execution on mobile platforms. Whether it is an octa-core x86 processor or a quad-core ARM processor, Spark should be the one unifying computing platform. We look forward to collaborating with the community to realize this effort, and hearing your feedback on the JIRA ticket: SPARK-6646.