[HARDWARE] Re: Alternative Options

Eno Jon enojon at delphi.com
Mon Sep 20 13:50:09 EDT 1999


Motherboards in a rack?   this pops up the query of a Beowulf
or WIndowsNT cluster running multiple instances of RC5 clients
sharing one or more buffer files (via a local proxy server?)

Something I've noticed with my "experimental" shared multiprocessors
(running two compatible versions of client under different os-es--albeit
one is i/o master)
is the file sharing oft means that one processor reads 2*2^28 block
while the other has begun processing an 8*2^28 block--partial blocks
either in the buffer itself or a result of the restart mechanisms?
The 2*2^28 block is then completed and the co-proc then reads
either a full or partial block.

I'm not sure what is occuring, anyone have a clue?


Jim C. Nasby wrote:

> Ooops... I guess I should have been more clear. By 'this kind of thing' below,
> I was reffering to a rack full of motherboards, not the FPGA project.
>
> Moo!
> dB!
>
> On Mon, Sep 20, 1999 at 08:57:39AM -0500, Daniel C. Petrak wrote:
> > Much of the cost involved [money and time] are in the research stages...
> > Seems like something well oriented for a group of people like EFF.  Anyone
> > have a contact with them?
> >
> > Regards,
> > Daniel C. Petrak
> > Senior System Administrator
> > The FCT Group of Companies
> >
> >
> > -----Original Message-----
> > From: Jim C. Nasby [mailto:jim at nasby.net]
> > Sent: Monday, September 20, 1999 1:49 AM
> > To: hardware at lists.distributed.net
> > Subject: Alternative Options (was: Re: [HARDWARE] my $0.02 re: dedicated
> > crack boxes)
> >
> >
> > I don't see much need for a coordinated project for this kind of thing,
> > unless someone wants to put the software into a nice easy to install
> > format. I'd venture that most of us on this list are very capable of
> > putting a bunch of motherboards into a rack.
> >
> > If you want to get really creative, see if you can figure out some way
> > to eliminate as much of the extra hardware as possible... use PPP via
> > the serial ports for example (especially usefull, since it gets rid of
> > the big NICs).
> >
> > As for some of the other options presented, they should be looked at,
> > but I know that we'll never get anywhere if a line isn't drawn somewhere
> > as to what the features will be. Yes, we'd obviously like fast, cheap,
> > and good, but you can only have 2 of the 3.
> >
> > There's already a lot of work done on the FPGA concept, and as complex
> > as it is, it's still _FAR_ simpler than trying another approach, such as
> > designing boards that would use some kind of regular micro running at a
> > high clock speed (where layout becomes *critical*).
> >
> > Just inserting my own $0.02 worth of reality.
> > dB!
> >
> > Paul Everett wrote:
> > >
> > > In my oppinion, I think that its still a better idea to just use some
> > > basic PC based machines for rc5 cracking.
> > >
> > > I'm all for building some kind of "rack of motherboards" as suggested
> > > somewhere here before.
> > >
> > > each unit would be:
> > >
> > > 2x celeron 400  106.00
> > > abit BP6 mobo    80.00
> > > 32mb dimm        41.00
> > > nic              12.00
> > > floppy            8.00
> > >
> > > (prices in pounds, excluding VAT, current from www.dabs.co.uk)
> > >
> > > this works out to about 185 US dollars.
> > > this would crack at about 2.2MKeys/second.
> > >
> > > You would need one PSU between 2 or 3 boards, adding about $10/board to
> > > the cost. So you have about $200/unit, including PSU and any oddments. You
> > > could also probably overclock the celerons a bit. If you managed to run
> > > then at (say) 500mhz, that would be about 2.75Mkeys/second, which works
> > > out to $73 per MKey/sec. Even without overclocking you have $91/MKey/sec.
> > > You could save a bit by getting rid of the floppy drive and using boot
> > > ROMs in the network cards. If you can overclock more than 500mhz (may be
> > > possible, i'm not an expert on this) then its even better value.
> > >
> > > This setup has several advantages over a FPGA/whatever solution:
> > >
> > > 1. The one off cost is low, there is no minimum order
> > > 2. the development costs are zero, as the hardware and software is
> > >    readily available.
> > > 3. as the machines are standard PC's, they could be used for other things
> > >    as well as rc5 - as rc5 runs with a low priority you could, say, use
> > >    them for rendering or compiling, but rc5 would use all the cpu when
> > >    you weren't doing anything else.
> > > 4. as the machines are standard PC's, you can sell them easily if you need
> > >    the cash or get bored of the project.
> > >
> > > And, as the prices of celeron processors drops (which it is doing) this
> > > becomes more and more preferable over other solutions.
> > >
> > > Whats more, I've actually done this. Well, not quite. About a year ago a
> > > built a rack pretty much as specified, but using 386 motherboards instead
> > > of dual celerons. I could easily just chuck some different motherboards in
> > > there though.
> > >
> > > Take a look at http://www.repton.org/racks/ if you want. Its got info on
> > > my rack, including some detail about exactly -how- i did it. There are
> > > photos too, and also details of a rack of Pentium 2 machines belonging to
> > > another d.net member.
> > > (p.s. if anyone else has done something similar, please send me info so i
> > > can add it)
> > > (p.p.s. if anyone from distributed.net is watching, and i know you are, i
> > > wouldn't say no to a link from the main site)
> > >
> > > Regards,
> > > Paul
> > >
> > > --
> > > Paul Everett                                 Microsoft is not the answer.
> > > repton at repton dot org                       Microsoft is the question.
> > > http://www.repton.org                                 "No" is the answer.
> > >    "God, root, what is the difference?" - Pitr, www.userfriendly.org
> > >
> > > --
> > > To unsubscribe, send 'unsubscribe hardware' to
> > majordomo at lists.distributed.net
> >
> > --
> > Jim C. Nasby (aka Decibel!)                                  /^\
> > jim at nasby.net                                               /___\
> > Freelance lighting designer and database developer         /  |  \
> > Member: Triangle Fraternity, Sports Car Club of America   /___|___\
> >
> > Give your computer some brain candy! www.distributed.net Team #1828
> > --
> > To unsubscribe, send 'unsubscribe hardware' to
> > majordomo at lists.distributed.net
> >
> > --
> > To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net
> >
>
> --
> Jim C. Nasby (aka Decibel!)                                  /^\
> jim at nasby.net                                               /___\
> Freelance lighting designer and database developer         /  |  \
> Member: Triangle Fraternity, Sports Car Club of America   /___|___\
>
> Give your computer some brain-candy! http://www.distributed.net Team #1828
> --
> To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net



--
To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net



More information about the Hardware mailing list