Hi Anish, thanks for keeping us posted about a progress!
CommonCrawl is important dataset and it would be awesome if we could find a way for you to build some notebooks for it though this this years GSoC program. How about running Zeppelin on a single big enough node in AWS for the sake of this notebook? If you use spot instance you could get even big instances for really affordable price of 2-4$ a day, just need to make sure your persist notebooks on S3 [1] to avoid loosing the data and shut down it for the night. AFAIK We do not have free any AWS credits for now, even for a GSoC students. If somebody knows a way to provide\get some - please feel free to chime in, I know there are some Amazonian people on the list :) But so far AWS spot instances is the most cost-effective solution I could imagine of. Bonus: if you host your instance in region us-east-1 - transfer from\to S3 will be free, as that's where CommonCrawl dataset is living. One more thing - please check out awesome WarcBase library [2] build by internet preservation community. I find it really helpful, working with web archives. On the notebook design: - to understand the context of this dataset better - please do some research how other people use it. What for, etc. Would be a great material for the blog post - try provide examples of all available formats: WARC, WET, WAT (in may be in same or different notebooks, it's up to you) - while using warcbase - mind that RDD persistence will not work until [3] is resolved, so avoid using if for now I understand that this can be a big task, so do not worry if that takes time (learning AWS, etc) - just keep us posted on your progress weekly and I'll be glad to help! 1. http://zeppelin.apache.org/docs/0.6.0-SNAPSHOT/storage/storage.html#notebook-storage-in-s3 2. https://github.com/lintool/warcbase 3. https://github.com/lintool/warcbase/issues/227 On Mon, Jul 4, 2016 at 7:00 PM, anish singh <[email protected]> wrote: > Hello, > > (everything outside Zeppelin) > I had started work on the common crawl datasets, and tried to first have a > look at only the data for May 2016. Out of the three formats available, I > chose the WET(plain text format). The data only for May is divided into > segments and there are 24492 such segments. I downloaded only the first > segment for May and got 432MB of data. Now the problem is that my laptop is > a very modest machine with core 2 duo processor and 3GB of RAM such that > even opening the downloaded data file in LibreWriter filled the RAM > completely and hung the machine and bringing the data directly into > zeppelin or analyzing it inside zeppelin seems impossible. As good as I > know, there are two ways in which I can proceed : > > 1) Buying a new laptop with more RAM and processor. OR > 2) Choosing another dataset > > I have no problem with either of the above ways or anything that you might > suggest but please let me know which way to proceed so that I may be able > to work in speed. Meanwhile, I will read more papers and publications on > possibilities of analyzing common crawl data. > > Thanks, > Anish.
