Topic: I found a way to implement real asic-resistant POW algorithm (Read 720 times)

You don't understand what I'm saying. You just spray trash

If you go to github, download and compile askcoin's source code, you will understand how I achieved ASIC-resistant

https://github.com/lichuan/askcoin

here is the ASIC-resistant data file:

https://github.com/lichuan/askcoin/blob/master/src/asic_resistant.data

The doc of askcoin's full node: https://github.com/lichuan/askcoin#askcoin

SHA256 + huge memory access, It can greatly increase the difficulty of asic mine.

The function described here is not asic resistant, in fact it looks like an outright scam. Buyer beware.

The mining process from askcoin project (www.askcoin.me) is:


The mainnet of askcoin is coming soon, the source code will be open when this project has passed its infancy.
FYI: https://bitcointalksearch.org/topic/annaskcoin-is-a-decentralized-real-time-qa-platform-3733771

There is a proof of work where the memory requirements can vary from 1KB to 1PB (a petabyte is a billion GB), where verification is instant, and where increasing the requirement is just a soft fork.

It is a MUST for at least two reasons:

1) because new nodes trying to identify the longest valid chain must process all proofs of work in the entire blockchain history. That's on the order of a million of them. Even if you allow an hour for downloading just the headers, that's only a few milliseconds you can spend on each pow.

and even more importantly:

2) a slow verification is an invitation for a Denial of Service attack where attackers keep feeding you bogus proofs-of-work which you need to spend significant time on to recognize as bogus. Verification should therefore take no more time than it takes to receive a header (well under a millisecond).

Why is it such a MUST?  Cumbersome validation, big deal.  Lets say it takes even 20 seconds to validate a block, a blocktime of 10 minutes means  that the head-start the original broadcasting miner gets is negligible.

And considering that the INVENTORS of Scrypt and Yescrypt say that they should have raised the N value speaks volumes to me.  Do you really think Charlie Lee or whoever's code he was copying knew how to implement Scrypt better for this application than the INVENTOR of Scrypt does?

There is a reason ZCash paid the inventor of Yescrypt to evaluate their use of Equihash.  Unfortunately they didn't heed his advice and now a short time period later they are suffering for it.

You're right, the real memory-hard algorithm should require that each data designed to be accessed from memory must be fetched from real memory. If some data could be deduced from other part by design some extra gate circuits, then it is not really memory-hard algorithm.

The N value is required to be kept low in Scrypt to allow costeffectiev and fast validation that is a MUST in PoW.

As I have mentioned above it is not about making algorithms gpu friendly. I guess you have been somhow misinformed about the situation. On the contrary, any PoW designer wishes to resists gpu optimization, it is just impractical imo.

The N value had to be set low so that verification wouldn't become cumbersome.


They chose a lower-than-desired memory setting because early solver implementations were comically inefficient and they worried miners would be too slow. They also lacked the patience to wait for miner improvements. Verification was never the issue.

Well lets look at scrypt for example.  Scrypt was tuned for GPU's with 1024 as the N value.  It was cracked by ASIC's.  If the N value was set to say 100,000 GPU's would likely struggle with it but CPU's would be able to do it pretty well.  To this day Scrypt would still be ASIC free in my view.  

Reading Solar Designer's (inventor of Yescrypt) analysis of ZCash's choice of Equihash, he also noted that they chose a too low "N value" type setting because they wanted it to be asymetric and verify very fast and be "GPU friendly".  It didn't work out too well.
https://blog.z.cash/the-zcash-equihash-analysis/

Here is his analysis of Argon2 Vs Yescrypt.  Seems Yescrypt is more parallelization resistant:
http://discussions.password-hashing.narkive.com/pYOIcGhK/phc-argon2-cpu-gpu-benchmarks


Good point, I didn't think about the Integrated Circuit part since everything is integrated circuits now.  But yes if they can integrate multiple parts of a typical motherboard in a way that excludes it's use in commodity hardware then that could be a definition of ASIC as well.

Although it is totally acceptable that always you can improve systems to be more adequate you can't do it in 'an order of magnitude' more efficient and cost effective way, at least you can't do it 'always'.

Plus, it is not just about tautology and playing with words. It is a god damned multi billion dollars business and our heads are right in the mouth of the lion. From a sociological point of view, it is highly recommended to be aware of the consequences of declaring ASICs as the ultimate winner of the PoW paradigm.

And even if  it was purely just a simple, innocent game about redefining and extending words I don't think it would be counted as a good move to extend Application Specific Integrated Circuit (what Bitmain is proud to be specialized in designing it) to include re-programmable appliances that are designed for a specific range of applications.

And there is very little difference between gpu and cpus in this context. Modern gpus are not optimized enough for I/O and typically are optimized for vector operations. PoW is naturally a good problem for gpus to solve and I don't believe it to be the inverse case, I mean designing algorithms for gpus.

Right, if the optimum "ASIC" that can be developed for a given algorithm -- if it's main components can all be directly used in "PC's" or general commodity hardware, then the "ASIC resistance" has succeeded in my view.

However we still have centralization issues like "less than order of magnitude improvements" -- joe blow still can't really profitably compete.  I believe the only algorithm that can solve this is "factorization of large numbers" which I invented.  That algorithm will require a GPU + a CPU and therefore all modern phones/pc's/laptops/AI will be able to mine that, whereas Server farms, Multiple CPU rigs, and Botnet's likely wouldn't (or at least wouldn't be able to gain a significant advantage).

https://bitcointalk.org/index.php?topic=2575256.0;all

To me that is really the point. A competitive single chip ASIC is a big hot-running chip requiring tons of design talent and access to the latest and greatest fabrication technology. This leads to manufacturing centralization and to noisy mining rigs that become obsolete within a year or two.

On the other hand, a chip that needs to interface to separate commodity memory chips can remain smaller, run cooler, and be implemented in a less than bleeding-edge technology, since it only needs to be fast enough to saturate the memory interface.


Yes, there is SRAM, but it still has a limited bandwidth.

I'm not just arguing semantics or definitions. It seems to me that the goal of "ASIC-resistance" is really to prevent a specialized device whose performance is an order of magnitude improvement over a PC, but at the same cost. Whether it is a single chip or not is not really the point, is it?

Now, I am not an EE so humor me, but aren't there RAMs that have access speeds that are an order of magnitude greater than standard DRAM, and wouldn't using that defeat the memory access bottleneck in an ASIC-resistant design?

I think he was just arguing semantics more than anything, just the definition of asic as application specific integrated circuit.  It is a tiring and unproductive argument though.  Technically I could make an asic for Yescrypt by removing usb ports and GPU ports and shrinking the board, maximizing copper for heat dissipation for 100% max CPU usage, etc.  The point is though it would still be a functioning computer, just designed for CPU mining.  So technically it would be an asic but not an asic by any real metric.  So basically ASIC just needs to be defined more closely to its real world form, say that to be defined as an asic, it must achieve at least an order of magnitude improvement in running a given application over general purpose computers.

But yes ASIC resistance is absolutely real; the problem is even algorithms that can be asic resistant are implemented in ways to benefit GPU's over CPU's. In this process of allowing for GPU acceleration, they make it significantly less resistant to real ASIC's.

Unbelievable question from a decent bitcointalk member 

Accessing memory bus is already optimized in cpu/gpu tech and it does not make sense for a  designer/manufacturer to claim such a thing: An asic optimized for memory bus access!

Any algorithm, (I mean any algorithm) being truly memory hard by means of memory access is inherently ASIC resistant as long it has two important properties:

• a large enough memory footprint
• a complex enough function that is not feasible to be implemented directly on the memory bank

The latter is new and I have to explain a bit:
It suggests that the fetch operation MUST be necessary and not feasible to be bypassed by adding some extra gates to a special memory bank to perform an in-place hash (like what happened for Equihash).

 The problem with Equihash and Cryptonight is that they have not achieved the desired level of hardness but Ethash (a Dagger Hasihmoto variant) is performing well and Bitmain's E3 by no means deserves to be called an ASIC.

To put it straight:
Any educated enough computer engineer can easily confirm that a memory bound algorithm can never be optimized by means of optimization of processing power (by implementing the algorithm in ASIC for instance).

So, it is just a ridiculous and naive assumption that every PoW algorithm is vulnerable to ASIC attacks.
On the contrary, I believe that ASIC resistance (and behind, ASIC proof) is achievable no matter what Bitmain and its agents try to induce in the community.

It need hard fork, but doesn't need any major code changes.

So are you saying this can be implemented in a soft fork?  Would there be any way for miners using old sha256 to fool the algorithm to accept their old type sha256 values, or skip a step of the memory hard part if their values look correct?  In other words, how is doing the memory hard part enforced?