TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

A Trillion Edge Graph on a Single Commodity Node

116 点作者 jcbeard大约 8 年前

5 条评论

thinkloop大约 8 年前
Isn't computing on a single machine generally always faster? Clustering is for handling more data than a single machine can hold, which requires paying network costs. There are parallelization benefits, but CPUs have so many cores now.
评论 #14224933 未加载
评论 #14222049 未加载
评论 #14222034 未加载
moab大约 8 年前
Very interesting paper, but the experimental section confuses me. They compare their system which preprocesses the graph into the tile-structure (a compressed representation), but none of the in-memory systems they compare against use a compressed representation. Is this comparison really apples-to-apples? If you can reduce the size in-memory of the graph by 60%, then I would expect an algorithm running on the compressed version to be faster than the same algorithm running on the uncompressed version (assuming that decoding the compressed adjacency-list is significantly cheaper than the cost of a cache-miss).
jacobparker大约 8 年前
Wasn't expecting to see the reference to the 9p file system!
douglasfshearer大约 8 年前
The paper doesn&#x27;t mention &quot;commodity&quot;, and I&#x27;m not sure that one to four Xeon Phi can be described as commodity.<p>Can the title be edited?
评论 #14222982 未加载
vertias718大约 8 年前
so, any repo code available yet?