Topic: [ANN][XEL] Elastic Project - The Decentralized Supercomputer - page 288. (Read 450524 times)

Thats a great point, if the values are low enough 40/60 might become trivial, and the costs savings for miners could be bigger than the difference, however it should not be custom (defined by job creators) as this could mean a power-struggle to the bottom (full PoW reward).

You can be sure that this very simple attack is one of the things which were thought of first  


Well, if you want to run your code on an infiniband cluster, you have to implement your code in OpenMPI,
if you want to run your code on parallelized GPU hardware, you have to implement your code in OpenCL or CUDA,
and if you want to run your code on the Elastic platform you have to implement your code in ElasticPL.

This is the nature of specialized platforms! Always was!

Right now ElasticPL is very basic, and i am sure that over time we will improve the instruction set and maybe even support other custom-built programming languages. Maybe ones which are tailored for specific use cases. See the current ElasticPL programming language as the first proof-of-concept which works for a certain type of programs. We can always extend what Elastic's tool set contains, but we will always have to stick to a custom programming language.

You cannot expect to support all types of programs, this will never be the case for elastic. We cannot allow users to upload arbitrary programs to the end user / miner's systems and execute it there. Neither can we apply any "proof of execution" check, nor can we guarantee that executing those programs is safe.

On the other hand, by using our own programming language like ElasticPL we can guarantee that execution of user provided functions IS safe. We exactly know the instruction set and can prove that there exists no code which accesses a user's disk/memory and that no code is capable of exeeding a certain amount of memory on the system.

Something as a compiler/translator could help.

This, unfortunately, severely limits the usability of the platform.  Most of codes, me, as a scientist, want to run are already existing software.

I understand why such decision was made, yet...

Just a thought:

Let's say I want a password bruteforced. What keeps a miner from sending the same hash over and over again? Are there entities in the network checking for that? If yes, how do they get paid?

I think there're too many factors (XEL price, miners' operating costs, computing power supply & demand, etc.) to make it static.
Scientists would probably be interested in fully customizable values.
More flexibility could make the "FLOP market" more competitive both for scientists and miners.

Actually OP should enhance the communication with us, add PM plz OP, and talk more with us. Thanks

How can someone get a hold of him with bitmessage?

Thanks.

5Mio$ is easily possible. And the coin amount is limited to only 5 Mio tokens. The run for the coin will be nice to watch.

I think the PoW/Bounty fund proportions should be dynamically calculated so that the reward per FLOP for PoW is always (significantly) smaller than the reward per FLOP of the bounty, that way there is another reason to not only mine PoW packets.

Thanks again for all the great work you're doing

Hi everyone,

First of all, sorry for my delayed response. Until Saturday I have very limited internet access.


I would say we are really really close to rolling out the testnet and start the bounty collection period for everyone who wants to hack around in Elastic's source code.

Before we can do so, we will just have to fix up all parts of the code marked with "FIXME". Right now that's 32 places in the code, most of them are either obsolete, already fixed or done very easily.

Additionally, but we can also do this while the code is being tested on the test net, is finish the live graph view showing the "efficiency" of the work package. The thing is, that I have no idea yet how to define "efficiency" ... maybe we should just show the "easimated number of flops" which somehow must be estimated from the program's worst case execution time (which in fact is the worst number of variable/register assignments), the current POW difficulty, and the rate at which new POW are generated. Any idea how we could eastimate that? Or any idea what we could plot to give he user an impression of how his program performs or if the interest in his program dropped (for example because other work packages were published that pay out more and who attracted the miners attention)?

Then, as we have many bright brains here, i would like to recite the current payout mechanism as it is implemented in Elastic. Any suggestions for improvement are welcome. So here is how it goes.

1. The scientist who wants some job done, programs his program in ElasticPL - our own programming language - and submits this work package to the network. Along with his transaction he puts a certain amount of XEL in escrow ... let's say 100 XEL.

2. 60% of it are put into the so-called Proof-of-Work fund, and 40% of it are put into the Bounty fund.

3. Now miners work on the ElasticPL program ... they basically bruteforce-search for inputs, which cause the ElasticPL program to evaluate to true. Most likely, the scientist is only interested in one or two inputs. If the scientist for example searches for a password which hashes to a certain hash, there will be most likely only one input which evaluates to true, this input is called a "bounty submission". Basically, a submission of something that the scientist is looking for. This is however not enough for a steady payout to the miners. Thats why we have also PoW submissions, that is inputs which hash to something else which meets a certain adjustable target value.

4. Every POW submission causes the payout of XXX XEL, a value the scientist can set himself and which makes his program more interesting on the market if the vaue is high enough. So "workers" who solve the scientist program get paid out steadily, even when they don't find the exact solution for what he is looking for. The work is closed when the POW fund runs out ... This is where the payouts come from.

5. Then we have the bounty fund, The bounty fund is not payed out upon every bounty submission. It is more like this, that bounties are collected as long as the maximum number of wanted bounties is found or if the work is closed/timeouted. Then, the bounty fund is distributed among all bounty submittors.

IT IS NOW VALID TO ASK, why do we not immediately pay out bounty submission the way we do it with POW submissions. Well, I thought about it for a very long time. Theoretically, it may be possible to just look for PoW solutions and not for bounties at all ... this can be done even faster by exploiting a variant of the Faster Algorithm Attack. This could cause people to earn money and not being ANY benefit for the scientist. A strong incentive had to be made ... with the payout afterwards anyone has the chance of being the one and only bounty solver and get the whole 40% of the bounty funds. This is a strong incentive. Also, if a few bounty submissions have been made, miners have also an incentive to continue working on finding bounties because each submision would increase their stash ... this is not the case with static outputs.

What we could now think about:
are the 60%/40% values O.K.? Should they be made dynamic instead of static. Is the incentive system with the chance of collecting the whole 40% fund enough for the users to care about submitting bounties instead of just going for the POW packages? Note ... POW submisions have NO VALUE to the scientist.

Second of all, and this could be Lannisters task, we need some programming bounties. I would suggest to outsource programming the faucet for the testnet users and offering some bounties for that? Also we could think about what we would need as well? The community is asked ;-)

Hi,

is there a Slack for this coin?

Thanks!

Perhaps it'd better to lower the fee ?
The supply will be increasing over time, so it's not neccessary to increase the initial supply

What lottery? Are you sober? NXT participants became milionares investing 2btc. Again, what lottery? Go be yourself elsewhere.

NXT ICO was totally a lottery, what do you think about this XEL?

When software will be ready. One month, three months, half a year. As soon as devs will make sure that they have ready and secure soft they will make a announcement of main net launch.

Mining period? XEL is PoS, NXT clone. What mining you're talking about?

When the mainnet will be active?

Will there be a mining period with XEL?

And you are one of the developers I assume?