Re: RAM for cluster net boot?

["Kevin D. Clark-4" <user=213@nabble.blu.org>]

Kristian Erik Hermansen writes: 

> Is there any usable distributed shared memory subsystem?  Even Google 
> doesn't do that, iirc.  Google's Map Reduce backend needs to split 
> jobs into finite operations that can be distributed across a network 
> of varying architectures and software.  The result is similar to what 
> a true DSM system would be, but without the complexity.  So, any 
> ideas?  Heh... 

Sure, there are lots of usable DSM subsystems.  I'm sure you can 
easily find quite a few. 

Of course, "usable" does not imply "this given DSM subsystem has such 
good performance that you can just use all memory as if it was local". 
In my experience, this fact has some pretty serious implications: in 
order to get even so-so performance, you have tune your code a lot. 
Really, a *lot*.  This is either (1) hard or (2) fantastically hard. 



Those Google guys mad skillz with MapReduce tell me that they are very 
smart and very practical... 

Regards, 

--kevin 
-- 
GnuPG ID: B280F24E             Don't you know there ain't no devil, 
alumni.unh.edu!kdc             there's just God when he's drunk? 
                                 -- Tom Waits 

-- 
This message has been scanned for viruses and 
dangerous content by MailScanner, and is 
believed to be clean. 

_______________________________________________ 
Discuss mailing list 
[hidden email] 
http://lists.blu.org/mailman/listinfo/discuss