Cluster Computing and MapReduce Lecture 3

Lecture 3: The Google File System. See
http://code.google.com/edu/content/su... for slides and other resources.

• Category

  Howto & Style

[Mike Beaubien]

I don't like his reasoning for why the reducers can't start until all mappers are done.

You can't just dismiss it by saying that you should just run more mappers instead. The fact is, you're limiting the opportunities you have for // execution.

Also, it wouldn't complicate the reduce code to have to deal with map failures. If a box fails during map, just kill all the reduce tasks that have received map input from it and restart them. An optimistic approach like this would perform well I think.

[ForexReviewsCentral]

a bit confusing...

[StroonsGlen]

please! ayaw kalimti ang slide, pahingi, gamiton nako sa ako classe.

[exn666]

бойан

[VondaFononono]

The Google File System is not that complicated as we thought.

[tivrfoa]

the link is broken

[Eido INOUE]

Thanks for the clarifying his exact words.

However, NFS isn't POSIX *compliant* either. Open Group's XNFS specs explain how NFS semantics differs from POSIX/SUS in over 10 pages of detail.

[Eido INOUE]

At the beginning of the lecture, he says that NFS is a POSIX file system. It isn't. I'm sure he just misspoke and meant to say that it's a networked file system popularly used with POSIX operating systems.