Mining centralization already happened but small blockers are afraid that the geographic area or region with fastest internet speeds will become the only place mining will be competitive if blocks get big. But I'm not buying their argument.
As far as the numbers, there's 8 bits in one byte. Therefore, a difference of 8 megabits in speed is one megabyte per second.
If block size is 8 MB, that's 8 seconds. Yet it takes 10 minutes to solve a block so 8/600.
Here's one of the small-blocker's argument regarding self-propagation as I understand it (I have
not rigorously gone through the math myself):
Large miners and mining pools have an advantage because the "propagation time" to their own hash power is fast compared to the propagation time to the rest of the hash power. Assuming worst case that their "self-propagation time" is zero, then larger miners and mining pools have two advantages:
1. Advantage due to latency (does not depend on block size). Recent work by G. Andrew Stone, suggests that average network latency for the propagation of block solutions is approximately 10 seconds.
This means that after a miner solves a block, he will have a 10 second "head start", meaning that he'll be hashing for
10 sec / 600 sec = 1.7%
longer than the other miners. This "advantage" occurs more frequently if you're a large miner or mining pool simply because you solve more blocks. If
h is the miner's hash power and
H is the network's hash power, the total advantage is equal to
1.7%
h/
H.
A miner with 20% of the network hash rate thus has a 1.7% x 20% =
0.34% advantage for winning the block reward. This advantage does not depend on block size.
2. Advantage due to bandwidth (does depend on block size)In the same paper, Andrew Stone also estimated the propagation impedance to be approximately 17 seconds / MB. Since miners can SPV mine on just the block header, the propagation impedance only affects their ability to claim the fees in a block.
Let's consider 2 MB blocks with 1 BTC in fees.
It will take
17 sec/MB x 2 MB = 34 sec
longer to propagate.
This means that after a miner solves a block, he will have a 34 second "head start" on claiming the fees in the block, meaning that he'll be hashing on a non-empty block for
34 sec / 600 sec = 5.7%
longer than the other miners. By the same rationale as before, his advantage with respect to the fees is
5.7%
h/
H.
A miner with 20% of the network hash rate thus has a 5.7% x 20% = 1.1% advantage for winning the fees. Since the fees were esimated as 1 BTC, this is a 1.1% / 25 =
0.044% advantage overall.
It seems to me that latency is more of an issue than the block size; however, both would appear to be negligible compared to things like the cost of electricity and availability of efficient hardware.