Momentum 2.0 Discussion

[barwizi]

I think 2.0 should be even more memory intensive and have a minimum processing threshold that has a large footprint process. By raising the minimum compute power required to hash we force botnets out of business by making it very noticeable to owners that their computer is being used. Also users of instances will have to hire out more expensive machines which may not give them ROI, i ROI.

with that in mind, let us always remember that there is no such thing as ASIC resistant algo, as long as the market has enough fiat supporting it and a growing requirement for more specialized machinery, one entity will venture to make one. So a more balanced approach should include a distribution plan that evens the curve in a manner beneficial to the community.

[bytemaster]

Would this go along with the vesting period?

Botnets, ASIC, and even to an extent cloud computing should be the overall concern.  If GPU mining provides a < 50% speed boost it still levels the playing field as both CPU and GPU's will get faster with new technology.  Also, most new cpgpu's can do opencl so I do not believe GPU's should be the ultimate enemy, however momentum algorithm may limit their effectiveness which is a plus.

Yes.  The goal was to make minor improvements in every area based upon lessons learned.  I agree that GPUs are not an enemy by themselves unless the reason the GPU successful is because the algorithm is trivial for an ASIC.   I believe this will keep ASICs and GPUs using a large amount of memory.

Any thought on lowering the vesting time period to under 6 months?  45-90 days would be nice 

Vesting is a separate discussion from the proof of work.   Lots of thought, but not sure how to evaluate the different periods so it is not set in stone.

[Brekyrself]

Would this go along with the vesting period?

Botnets, ASIC, and even to an extent cloud computing should be the overall concern.  If GPU mining provides a < 50% speed boost it still levels the playing field as both CPU and GPU's will get faster with new technology.  Also, most new cpgpu's can do opencl so I do not believe GPU's should be the ultimate enemy, however momentum algorithm may limit their effectiveness which is a plus.

Yes.  The goal was to make minor improvements in every area based upon lessons learned.  I agree that GPUs are not an enemy by themselves unless the reason the GPU successful is because the algorithm is trivial for an ASIC.   I believe this will keep ASICs and GPUs using a large amount of memory.

Any thought on lowering the vesting time period to under 6 months?  45-90 days would be nice 

[bytemaster]

Would this go along with the vesting period?

Botnets, ASIC, and even to an extent cloud computing should be the overall concern.  If GPU mining provides a < 50% speed boost it still levels the playing field as both CPU and GPU's will get faster with new technology.  Also, most new cpgpu's can do opencl so I do not believe GPU's should be the ultimate enemy, however momentum algorithm may limit their effectiveness which is a plus.

Yes.  The goal was to make minor improvements in every area based upon lessons learned.  I agree that GPUs are not an enemy by themselves unless the reason the GPU successful is because the algorithm is trivial for an ASIC.   I believe this will keep ASICs and GPUs using a large amount of memory.

[Brekyrself]

Would this go along with the vesting period?

Botnets, ASIC, and even to an extent cloud computing should be the overall concern.  If GPU mining provides a < 50% speed boost it still levels the playing field as both CPU and GPU's will get faster with new technology.  Also, most new cpgpu's can do opencl so I do not believe GPU's should be the ultimate enemy, however momentum algorithm may limit their effectiveness which is a plus.

[bytemaster]

I have been thinking about ways to restore Momentum to a memory-hard problem and have had an insight that I believe changes the game.

Lets define a Birthday to be 1 MB of 'random data' that is relatively expensive to calculate and seeded with a nonce
Lets define a Collision to be the number of bits in common between two random sequences. 
The goal is to produce two nonces such that the number of bits in common between the two 1 MB sequences is above some threshold.
The cost of validation is populating 2 MB of memory, performing an XOR, and then a population count.
Now lets populate the memory with AES which has hardware acceleration on CPU but not GPU

The population of the memory is sequential which helps eliminate memory bus latency and allows the CPU to saturate the BUS.  A GPU would have an advantage in the comparison step because it could compare all birthdays with the new birthday in parallel.    That said, a GPU would be limited in parallelism because it could only store 1000 or 2000 birthdays at a time.   Even so I am not convinced that a GPU implementation would be a bad thing given the other changes we are making.

This proof is slightly harder to validate, but most of the optimization techniques used to accelerate Momentum 1.0 would not scale as well with this system and an ASIC would still require a very large amount of memory.   

What do you all have to say?