sync-ing two Props.
rjo_
Posts: 1,825
We all know that the Prop is a very deterministic device. The question naturally occurs, "how deterministic?"
Has anyone done the equivalent of the following:
1. hook up two props to the same crystal,
2. place identicle code on both eeproms, which code outputs a signal to same numbered pins,
3. force simultaneous resets, and then
4. monitor the sync between the pins?
Of course the sync should be perfect... but there are all kinds of things that might cause variable differences.
If in fact this kind of sync can be relied upon, then it would seem to me that the Prop is unique in an important aspect.
Rich
Has anyone done the equivalent of the following:
1. hook up two props to the same crystal,
2. place identicle code on both eeproms, which code outputs a signal to same numbered pins,
3. force simultaneous resets, and then
4. monitor the sync between the pins?
Of course the sync should be perfect... but there are all kinds of things that might cause variable differences.
If in fact this kind of sync can be relied upon, then it would seem to me that the Prop is unique in an important aspect.
Rich
Comments
1) The PLL's (being analog) could start up a bit differently in the two devices. Once started the propellers should remain in step with each other - there being only a phase difference. Chip could probably tell you what the range of variation is. The impression I got was that it wasn't big enough to use for random number generation - being not random enough.
2) Variations in input threshhold could result in an identical signal being seen at slightly different times between two propellers.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
You have an almost mystical ability to get to the point.
Thanks
Rich
I would very much enjoy reading a technical discussion between experts on this point. How unique is this capacity?... how might it reduce overhead in a parallel implementations?... what is the pertinent history, ... etc.
(3/4) This sync is not possible, as the Pros run for a considerable time after reset with the internal RC clock, which differs from chip to chip.
WRT your last question. Other processors are by no means less deterministic than the Prop. "Determinism" is generally destroyed by software: The slightest IF will put an end to it.
Chips with one processor only are handicapped by the necessity for interrupt routines for any non-trivial operation.
The interesting feature of the Prop - IMHO - is that the concept of "trivial operation" goes a long way with eight COGs
Thanks for your response.
In Chicago, you would be called "da" Silva.
I had to look up the concept of "trivial operations." In terms of the significance of trivial operations, I found: http://www.ece.uvic.ca/~amirali/publications/vpw2.pdf
The significance seems to be tha one can reliably anticipate and thereby reduce code at the pre-processor level... is this what you are saying is a major strength of the Propeller? Could you elaborate a little?
By the way, RadioShack in the United States would appear to be in need of a consultant to help their managers[noparse]:)[/noparse]
Talk to the CEO's executive secretary. In the US... the women run the show. Try to be charming.
Rich
I was using the term "trivial operations" in a looser sense: I meant a sequence or a loop of instructions without decisions - or very simple decisions - inside, e.g. to break the loop.
Those kinds of "programs" were typical for early computers, as the Zuse machines, reading the program from paper tape, or the ENIAC, which used hard-wired code.
Such programs are generally easy to write and to understand.
Complexity is introduced in many ways into programs, the best known is "recursion". For "real time" machines, collateral execution of code strings is more typical (Perhaps someone will remember the discussion along a Djkstra paper, in which he explains the context of his invention of "semaphores")
The hardware support for parallel execution is so splendid on the Propeller, as it needs neither operating system interaction, nor other considerations from the programmer (as e.g. co-operative multitasking does).
Post Edited (deSilva) : 10/15/2007 8:23:38 PM GMT
Thanks.
Rich
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
The more I know, the more I know I don't know.· Is this what they call Wisdom?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
The more I know, the more I know I don't know.· Is this what they call Wisdom?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
The more I know, the more I know I don't know.· Is this what they call Wisdom?
However,if that clock is not synchronized to the reset it might be too early for one the chips, so tying it to the (master-) reset will be a good idea I think.
Nevertheless a CYCLICAL soft sync (1 Sec?) will be even better.
I guess the original question was related to have two props in "perfect" sync, which may not be possible without a way to re-sync the hubs.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
The more I know, the more I know I don't know.· Is this what they call Wisdom?
Assuming a single clock and multiple chips, then (worst case) you would end up with a "fixed" coherence length determined by the hubs and possibly a variable phase angle, determined by the cogs.
Of course what everyone would want is an infinite coherence length and a zero phase angle... but any predictable relationship would do.
Some important real systems, which one might want to measure, are sometimes governed soley by changes in coherence and phase angle. For example, one very common system that I know of would require a minimum of 5 rotating frames per locus to fully rationalize the data. Simply following the phase angle and coherence length of any one of these 5 frames gives you a lot of information. Usually, the first step is to physically impose some order on these frames. The read out rate is determined by this first step and the choice of spin systems.
This is not my primary interest, as it carries a lot of baggage, but there is a huge market that spans most of the basic sciences. Convincing a ranking physical chemist to look at the lab value of the Prop architecture makes perfect sense to me.
Rich
One correction... in the United States and England, these sorts of applications would fall within the domain of physical chemistry. In Russia and parts of the old Eastern bloc, the same applications are handled by biophysicists. Some targeted marketing would go a long way to placing the Prop into several instrument markets.
Rich
And one more opinion: if one had intentionally set out to design a chip, which would ideally serve some of these critical markets, one could not have done better than to design a Propeller.
Rich