Quote from: ABitNut on February 05, 2015, 09:20:05 AM

Quote from: johnyj on February 05, 2015, 08:32:25 AM

You must be able to broadcast that huge block to most of the nodes in 10 minutes. I don't see the latest research regarding this area, but there is a paper from 2013

http://www.tik.ee.ethz.ch/file/49318d3f56c1d525aabf7fda78b23fc0/P2P2013_041.pdf

Based on this research, it took 0.25 seconds for each KB transaction to reach 90% of network. In another word, a 1MB block will take 256 seconds to broadcast to majority of nodes and that is 4 minutes

When block size reach 10MB, you will have a broadcast time of 40 minutes, means before your block reach the far end of the network, those nodes have already digged out 3 extra blocks thus your block is always orphaned by them. And the whole network will have disagreement about which segment have the longest chain, thus fork into different chains

Gavin's proposal is to let mining pools and farms connect to high speed nodes on internet backbone. That is reasonable, since the propagation time is only meaningful for miners, your transaction will be picked up by the mining nodes closest to you and if those mining nodes have enough bandwidth, they can keep up with the speed. But anyway, how much bandwidth is really needed to broadcast 10MB message in a couple of minutes between hundreds of high speed nodes need to be tested. And this is the risk that someone worried about the centralization of mining nodes: Only those who have ultra high speed internet connection can act as nodes (I'm afraid that chinese farms will be dropped out since their connection to the outside world is extremely slow, they will just fork to their own chain inside mainland china)

I don't know how you come to those assumptions based on that research.

Quote

the block message may be very large up to 500kB at the time of writing.

Quote

The median time until a node receives a block is 6.5 seconds whereas the mean is at 12.6 seconds.

Quote

For blocks, whose size is larger than 20kB, each kilobyte in size costs an additional 80ms delay until a majority knows about the block.

The do not mention the average size of blocks they measured. Let's assume all their blocks were 0KB. 12.6 seconds for that. Add 80 ms per addicition KB.... 80ms * 1024 * 20 is about 27.3 minutes. Add the original 12.6 seconds... Roughly 28 minutes for 20MB.

Of course, 28 minutes is still long. That is based on 2013 data. I assume the nodes now will have improved their verification speed and have more bandwidth. New measurements could / should be made to verify that propagation speed will not become an issue.

I just took the numbers on that chart, their paper says 0.08s/KB but the chart shows 0.25s/KB, no big difference

Ideally, you would like to keep the broadcasting time below 1 minute, to make sure the network does not fork into different chains, and to reduce the orphaned blocks. Currently some connection to china is about 20KB/second, means 1MB data will take 1 minute to just reach their network. Of course china have much less nodes than the rest of the world, but they do have large amount of hashing power

For 10MB block, the bandwidth requirement will be 2Mb, which is quite high if you consider the connection over continent. Hopefully before we reach that stage the network bandwidth has been upgraded