The dream platforms for Dragonfly?

Andreas Hauser andy at splashground.de
Fri Sep 24 02:59:35 PDT 2004


dillon wrote @ Thu, 23 Sep 2004 22:18:19 -0700 (PDT):
> 
> :So, is this the kind of machine that Dragonfly is hoping to make
> :the most of?
> :
> :http://www.orionmulti.com/news/pr_083020041
> :
> :(one product is a 12-cpu cluster in a desktop box, and the other
> :is a 96-cpu cluster in a somewhat bigger box.  Transmeta CPU's
> :to keep the power requirements down...)
> :
> :-- 
> :Garance Alistair Drosehn            =   gad at xxxxxxxxxxxxxxxxxxxx
> :Senior Systems Programmer           or  gad at xxxxxxxxxxx
> 
>    The clustering scheme that I am most intent on supporting is a LAN or
>    WAN based cluster of discrete machines.  Think of, oh, setiathome running
>    on tens of thousands of machines that look like one logical machine. 
>    At the extreme end, of course.

[Before i get another bunch of useless replies from people that
missunderstand my intentions, let me stress that DF for me is one
of the greatest things that happened in this millenium until now
(yes i live in a small world)]

I've built OpenMosix clusters and really can support the great user
interface of SSI  in general (but UNIX management interfaces are not
mighty enough to really deal with that new world).
But isn't it a bit early to buy hardware for that kind of stuff?
And some other parts of your vision aren't even solved, i think.
Let alone implemented.
How are you thinking you solve the latency issues (and QoS) on a WAN e.g.?
A lot of experiments and research will be burned on a scheduler that
can do these things on a fast reliable LAN alone.
Another issue that needs imo the attention is that one needs to replace
the (long aged) UNIX system resource and rights management.
I mean it doesn't even solve the problems we have now (You can't e.g.
restrain your users from taking you down with a fork bomb in a sane way.)
After that you keep contact to UNIX in a compatibility layer at max.
And then you are in open land and safe pathes will be harder and harder
to find.
Dunno but i get a wierd feeling when ppl start asking what HW they shall
buy for something i imagine like 5 years away and with only a certain
probability of success. So i think the _long time_ part of the project
should be more stressed and for the short term the usual hardware
should do.


Andy





More information about the Kernel mailing list