Hacker News new | ask | show | jobs
by Anonobread 3811 days ago
> In any case, fixing the blocksize limit will be good

This presumes the blocksize limit is being fixed, which is an incorrect assumption. Originally, the proponents of Classic wanted to boost block size limits to 20MB effective January 2016 [1], and used bombastic and divisive language, inventing their own crisis even, to get people to take them seriously.

Interestingly, those same people are now contending for just 2MB, which is only 0.4MB higher than that proposed by Core.

Performance test data shows 32MB is the absolute maximum block size you can handle on a modern desktop PC today [2]. And by this I mean you could validate no more than one block per 10 minutes, hence realistically you actually couldn't run a full node on a desktop at 32MB. You'd need a clustered full node just to keep up, which is unprecedented in Bitcoin.

To make matters worse, 32MB blocks gets Bitcoin to 300 tps which is less than 1% of VISA's current capacity.

[1]: http://gavinandresen.ninja/does-more-transactions-necessaril...

[2]: https://blog.conformal.com/btcsim-simulating-the-rise-of-bit...

4 comments

> Interestingly, those same people are now contending for just 2MB...

Hearn mentions the backstory behind this in a comment here [0], posted a little while after you posted your comment. After reading his comment, does the shift from 20MB max blocks to something smaller become less "interesting" and more understandable? If not, why not?

Edit: Additionally, do you have anything to say about this comment? [1]

[0] https://news.ycombinator.com/item?id=10921219

[1] https://news.ycombinator.com/item?id=10921209

Well, any improvement to blocksize will be good, and will hopefully break the logjam preventing changes.

If we wait for a perfect fix to all problems, current and projected, we'll be waiting forever.

>Performance test data shows 32MB is the absolute maximum block size you can handle on a modern desktop PC today [2].

No, it's maximum size you can handle using a single-threaded poorly-optimized cpu software.

>Additionally, a profile of the CPU usage of this node, using golang’s great profile capabilities, shows that the CPU usage is dominated by the ECDSA signature verifications

OpenCL on gpu can easily process 50 MILLION bitcoin addresses per second (ecdsa + sha). [0] Note that OpenCL is available even on smartphones. For a very rough estimate, Adreno 530 on Snapdragon 820 has 500GFlops, 10% of 780 ti which can do 50M/s, so it should do about 5M/s. So a $400 smartphone should be more than enough to process 32MB blocks - provided the gpu is used.

[0] https://en.bitcoin.it/wiki/Vanitygen

This... does not sound good at all. Frankly, I wasn't expecting that scaling the block size would be so taxing on normal nodes.