shall we write our own? Re: [Beowulf] O'Reilly Clusters Book Review

[Mark C. Ballew]

On Fri, 2005-02-25 at 18:33 +0100, Ryan Sweet wrote:
> Glenn,
> 
> I have also had a look at the new ORA cluster book, hoping that they had 
> learned their lesson, and had a similar reaction.  While I don't wish to 
> discredit M. Sloan, as writing any sort of book is always going to be a lot of 
> work and filled with compromise, I felt from the very beginning that the 
> community can do better.  After reading over your review, which, while 
> scathing, was entirely accurate, I feel resolved that the beowulf community 
> _should_ do better.
> 
> Here's what I propose: let's make a "Beowulf.org Guide to Linux Clustering", 
> or whatever the heck else you want to call it.   Let us outline, review, 
> improve, and comment on it here on this mailing list.  Here's the hard part - 
> lets also set a deadline, with realistic goals, and try to stick to it.  Lets 
> assign any publishing rights or other "details" like that to the FSF or Linux 
> Documentation Project.
> 
> Robert Brown has already done a lot of work on such a book, and generously 
> made it freely available.  Maybe he is amenable to this being a starting 
> point?
> 
> In any case, I would gladly provide hosting for something like this, and 
> coordinate the project, as well as edit or write content.
> 
> There are many questions that arise:
>  	Most importantly - What should go in the book?
>  	In what order should these topics be covered?
>  	Should there be an attempt to have a common style?
>  	How (and how often) should it be revised?
>  	Does the book target new beowulf admins, seasoned experts, or both and 
> some in-between?
>  	Should mentioning vendors be allowed? What are the guidelines?
> 
> and so on.
> 
> Firstly, now that I've proposed the idea, I'll also start by volunteering to 
> write a chapter on diskless clustering.
> 
> Second, please take this opportunity to tell me why this is a bad idea, and 
> while your at it send your comments on the questions above.

I think a community-written book is a spectacular idea. The question I
have is would it be better to just do a web-based book since the beowulf
community is basically a moving target, or just put these into printed
"editions" as well as a Copyleft'd book?

I volunteer for any editing or proofing if such a project comes to life.

What goes in the book? Types of clusters, cluster purposes, cluster
interconnects, common issues (HVAC, user admin), and scaling are
somethings that comes to mind.

Order? Start with the basics. What is a cluster? Why clusters and not
SMP beasts? What types are there? What software is there? 

Style: Good question

How often should it be revised? If it is web and dead tree based, the
web version would constantly be updated with perhaps a yearly dead tree
revision?

Who does it target? I think a book that targeted beowulf admins with
some stuff for the seasoned expert would be good. New beowulf admins
often find themselves swamped with options. Seasoned admins join the
mailing list.

Vendors: I think that vendors should be allowed but only if there is a
balance between free options and vendors, or in the case of hardware,
why you'd go with a particular vendor (Myrinet vs. Infiniband, etc.). I
think something like "you should go to PSSC or Penguin Computing" is a
little bit of a stretch beyond an appendix on various vendors.

Mark
-- 
Mark C. Ballew                          Reno, Nevada    
ballew at sublinear.net                    http://markballew.com
PGP: 0xB2A33008                         AIM: pdx110
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://www.scyld.com/pipermail/beowulf/attachments/20050225/41cd413c/attachment.bin