[Beowulf] large MPI adopters

[Vincent Diepeveen]

On Oct 8, 2008, at 4:09 PM, Bogdan Costescu wrote:

> On Tue, 7 Oct 2008, Vincent Diepeveen wrote:
>
>> Drug type companies eat typical 0.5% system time.
>
> Could you please explain what you mean here ? That they buy lots of  
> CPUs but only use a fraction of them or only for a fraction of time ?
>

When you add up others.

Note i see in IWR you got a 160 node opteron cluster, what software  
eats system time on that cluster?
There must be graphs about it, want to share one? Also interesting to  
know is how many cores most jobs eat!

Note that i left out 1 important factor which is not a factor in  
every nation: that's the amount of system time airindustry eats of  
supercomputers
designing new stuff. It is very significant together with car  
manufacturers. With 160 nodes machine, even if that would be power6  
nodes,
you can't even design a wing for aircraft.

>> They are far behind, which is not so logical; i would consider it  
>> old fashioned to do experiments on monkeys (Bar-ilan brain  
>> experiments on monkeys rings a bell?), so doing everything in  
>> software is far more logical.
>
> Maybe your knowledge about the pharma industry is far behind ?  
> "Doing everything in software is far more logical" only when you  
> can exactly express biological processes in mathematical terms  
> which is not true in the great majority of cases. F.e. molecular  
> dynamics (often used to simulate interactions at molecular level,  
> f.e. of a drug injected into the body) is based on several  
> approximations and the implementations are subject to numerical  
> errors. Would you take a drug that was produced based only on a  
> simulation with molecular dynamics and found to be good enough ?
>

Well, i sure avoid hospital visits, you?

As you know the healthcare commission is basically old men who spread  
careful chosen sentences in such a manner that
everything looks fine on this planet and that there is nothing to  
worry about; last time i spoke with them on EMF's a guy being spokesman
on this subject, he didn't even know how the electricity system in  
his own house worked (whether on the first floor of his house
he had earth).

So lucky to introduce to a new drug, there is some burocratic  
systems. There is several phases of testing that a drug
has to pass in order to get accepted as a drug that is legal. Not to  
confuse with the 'advice' on whether a drug gets prescribed;
they want way cheaper drugs for that usually than the ones that might  
maybe be better (say 0.5% better for a lot of money extra,
but if you're that guy that didn't get cured as a result of that,  
your family will regret it bigtime).

Computing power or not, most manufacturers simply want to sell a  
product, so the amount of systemtime invested on a computer,
in order to sell, is total irrelevant to them to get a drug accepted  
usually. The first and most important part of the show to get a drug
accepted is getting invited to the table to show your drug. Computing  
power is nice to mention here when that is usually not a factor
at all to get a new medicine accepted.

But it may change, let's hope so...

Maybe a new breed of company can do the job. A breed where making  
profit is not most important :)

> Similar concerns apply to all levels - the brain, as you mention  
> it, is still very much an enigma, otherwise artificial intelligence  
> would be all around us (or not, if deemed to dangerous...). And  
> when you don't know how an organ or system works, how can you  
> develops drugs only by simulations ?
>
>> When someone is really in big need for big calculation power, one  
>> makes dedicated cpu's for it.
>
> I tend to agree, provided that the dedicated CPU brings a really  
> big advantage, like an order of more of magnitude. F.e., to stay in  
> the pharma related field, that's the bussiness case for DEShaw  
> Research's molecular dynamics chip (Anton) which is supposed to  
> bring a significant increase in speed for MD simulations compared  
> with software solutions running on general purpose CPUs (or GPUs).

Some hardware guys can better comment here. Making your own FPGA  
chippie is rather cheap.
Finding and paying a few good guys to get the chip done is toughest.  
You basically need 1 good
programmer, the rest can get hired parttime in fact. After chip works,
to print a 1000 cards, it is like a 100 dollar for pci-card and 50  
dollar a chip.

So $150k for computing power that total annihilates any supercomputer  
of now and the NEAR future for your specific application.

Real cpu's are a lot faster of course, price for them i do not have  
at hand. I heard it could be cheaper
than this in fact if you regurarly print a batch. Such a batch is  
producing less cpu's than the FPGA, where you pay for 1 cpu a constant
amount. I tend to remember that good cpu designers, not to mention  
memory controller guys, are rather expensive folks, as
they are very popular; on some other mailing list with some very  
funny lastnames, i see nonstop job offers for that type of folks.

Vincent


> -- 
> Bogdan Costescu
>
> IWR, University of Heidelberg, INF 368, D-69120 Heidelberg, Germany
> Phone: +49 6221 54 8869/8240, Fax: +49 6221 54 8868/8850
> E-mail: bogdan.costescu at iwr.uni-heidelberg.de
>