I don‘t think this is really do to the skew in the data. 100k is still a very small amount of messages. It just means that, that vertex will likely generate or receive 100k * 8 = 800k bytes = 0.8 MB more data. How much memory are you giving to your java virtual machine? How much memory do your machines have. I have about 4GB on each of my machines so I give the following flags to my java scripts: -Xmx3000M. You should change the script file here: https://www.assembla.com/code/phd-projects/subversion/nodes/gps/trunk/scripts/start_gps_node.sh?rev=95 There are two jvm flags XMX_SIZE and XMS_SIZE, which you should adjust.

　　于是乎我就跑过去修改了一下，分别增大XMX_SIZE 和 XMS_SIZE，然后又出现新的问题：连接不上端口，于是重新修改端口号，在传到hdfs上面去。

　　好了！

 

 

 

2.

第二个问题，下午跑RDFlib的时候（我主要用它来使得解析RDF文件，使其变成图数据）。一开始使用 easy_install rdflib来安装的时候总是显示有错误（可能这几天网络有问题，梯子不够长，嘿嘿）。后来急了，直接下载源文件，本地手动安装！可是找了半天，居然没有找到怎么手动安装！！剁手～～～其实，进去文件目录后，prthon setup.py install就可以了。

安装好了，尝试跑数据了，小数据跑得呼呼爽，后来准备跑真实数据了，300M 左右，结果就怂了。

No handlers could be found for logger "rdflib.term"

后来搜了一下，这个网址里面有解决方案 http://stackoverflow.com/questions/17393664/no-handlers-could-be-found-for-logger-rdflib-term

import logging
import rdflib

logging.basicConfig()# now load your graph
g = rdflib.Graph()
g.load("life_the_universe_everything.rdf")

3.

跑rdflib的时候，遇到问题：
WARNING:rdflib.term:http://www.w3.org/1999/02/22-rdf-syntax-ns# first  does not look like a valid URI, trying to serialize this will break.

然后会直接导致不能运行，郁闷啊！况且我的代码怎么能有Warning！于是乎去改，搜索了一下啊，有人说把URL里面的空格岁百纳用什么替代就好了，瞬间就笑开颜了，哈哈，果然好用！
当然了，我这么操作是因为我不在意具体的URL是什么，我只是把它当作一串字符而已！