[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Reading list



> The great thing about this type of question is that it can easily 
> be the start of a nice little flame war.  :)

Indeed =)

> I have found, however, that books are sometimes not a very helpful 
> source of information in this particular field.  

I'll add to this. I've read Becker's own "How to Build a Beowulf", 
and can say he's right on the point when he mentions that "all 
litterature about clusters is hopelessly outdated by the time it 
comes out" (or words to that effect).

Nothing can beat having a knowledgeable linux admin handy, some free 
time, and an awful lot of patience (and an understanding boss, 
depending...)

What you should clue yourself about is basic multi-threading 
concepts and networking issues, and then try to run (and understand) 
a few cluster-specific softwares.

In my case, the 'training' was 'install it and play with it 'til it 
works'. Seems to have worked (in a kinda-sort-of way).

Another big problem is that most clusters are custom-designed for a 
specific task. In our case, we ended up with an hybrid machine 
running OpenPBS (queue), MOSIX and LAM/MPI custom software on a 
cluster that is our new departmental server for processing of all 
sorts. Unless you are clustering for home use (in which case you 
should try anything you can put your hand on), expect your 
requirements to vary wildly from mine, from the next guy's, etc.

Wildly enough to invalidate any simple HOWTO or book on clustering. 
You WILL need to educate yourself, to learn what you need it to do, 
to investigate available software, etc.

Christian Lavoie
clavoie@bmed.mcgill.ca
0

Reply to: