ts-7000
[Top] [All Lists]

[ts-7000] Re: 16 Node TS-7200 Cluster

To:
Subject: [ts-7000] Re: 16 Node TS-7200 Cluster
From: "Jesse Off" <>
Date: Tue, 25 Oct 2005 03:14:45 -0000
Congratulations and thanks Ron for the great article!  Our office was 
quite excited (and surprised!) to see our boards in the Linux Journal 
in such an interesting application.  We've always toyed with the idea 
of productively using our TS-72xx burn-in racks as a sort of 
distributed cluster.  Its fun to know that someone has actually done 
it!

Interesting idea about the "link boards", but likely the onchip 
ethernet is the fastest communication bus available for these 
boards.  100 Mbps on a 100Mhz internal-to-the-processor AHB bus with 
internal DMA access to SDRAM is hard to beat on this class of 
processor.  Even the cheapest of 8-port ethernet switches typically 
have at least a 4Gbps cross-switched backplane to shove and broadcast 
packets around.  If you can keep the software driver and network 
stack overhead low,  ethernet is a great high speed message passing 
interface.

The EP93xx ethernet driver could be a lot faster if Linux could 
guarantee 32-bit aligned packet storage.  Currently, if a packet 
happens to not be, the CPU ends up copying the entire packet to 4-
byte aligned storage.  I'm not sure how often this happens, but the 
fact that Linux can maintain about 80-90Mbps on low-overhead UDP 
despite this is pretty impressive IMHO.  To acheive this kind of raw 
bandwidth on the PC104 bus you'd be running faster than 200nS 
cycles.  You can get this speed from the TS-7200 PC/104, but you have 
to re-program the SMCBCR timing registers from their default settings.

Ron, one of our production employees has shown me that we have some 
metal machined stand-offs that may have saved you some time in 
stacking TS-7200's rather than making chain-mail spacer rings (which 
was rather creative, BTW!) 

Also, you left it open in your article as to whether you needed 
the "toolbox fan" to cool the 16 TS-7200s. I will attempt to 
comment.  16 TS-7200's will generate about 32 watts of heat, or about 
half what a 60 watt light-bulb puts out.  We have a temperature 
chamber where we regularly test production boards to 80-90 degC.  80-
90 degC is very hot to the touch and while the thermal resistance of 
your plastic toolbox is worse than a metal one would be, I would be 
very surprised if 32 watts of heat could generate 80 degC ambient 
temperatures inside the toolbox even without the case-fan.

//Jesse Off   



--- In  "vocemanago" <> wrote:
>
> --- In  Ronald G Minnich <> 
wrote:
> >
> > dbpalan wrote:
> 
> > 
> > the connection is really simple. It is ethernet. The power is the 
> > straight 2-wire 5V connection from the Parvus power supply.
> > 
> > Sorry, no 2.6 port yet.
> > 
> > Due to the lack of a bridge between the CPU and PC/104, we can't 
just 
> > stack the PC/104+ busses; we'd like to, however.
> > 
> > thanks
> > 
> > ron
> >
> I could see a neato cross link between CPUs utilizing an card that 
had
> an epld on it to connect one 16 bit bus to the other. 
> It would be processor, link board, processor, link board, processor.
> The link board would have a surface mount connector on each side so
> they wouldn't actually be connected together. The EPLD would be
> between them. The connectors wouldn't be that strong but once you
> build the stack.
> Then each board from the bottom would talk to the one above through 
an
> 8 bit register and and interrupt, and each board from the top would
> have a reverse register. 
> The question would then be would a 100mbs serial bus and its
> associated overhead be faster than the pc/104 parallel bus. 
> A driver would have to be created , possibly some bastardised 
version
> of plip used for parallel port comunications.
> 
> Anyways, its an interesting toy to think about.
> Tony
>






------------------------ Yahoo! Groups Sponsor --------------------~--> 
Fair play? Video games influencing politics. Click and talk back!
http://us.click.yahoo.com/T8sf5C/tzNLAA/TtwFAA/CFFolB/TM
--------------------------------------------------------------------~-> 

 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/ts-7000/

<*> To unsubscribe from this group, send an email to:
    

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/
 


<Prev in Thread] Current Thread [Next in Thread>
Admin

Disclaimer: Neither Andrew Taylor nor the University of NSW School of Computer and Engineering take any responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the birding-aus mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU