2016 blocks = 14 days = 10 minutes per block - or not?

[shamoons]

So I was wondering and perhaps the technical forum is a better place for this discussion. But the network self-adjusts every 2016 blocks to set the difficulty to be equal to what it should have been to ensure that 2016 blocks took 14 days. So if there's more network power in a 2-week period, the 2016 blocks will take less than 14 days. The network will then adjust it so that the difficulty is harder so that the next 2016 blocks (at the current network power) takes 14 days. But the problem is that we already had the first set of 2016 blocks take less than 14 days. So in time, the average time to find a block is less than 10 minutes.

Am I missing something?

[Remember remember the 5th of November]

Block finding is random. I have seen blocks being found within seconds of each other. It all depends on luck. Sometimes due to luck, blocks are found faster even if the difficulty corresponds to the network computational power.
And sometimes, again due to luck, some blocks take a while(20-30 minutes).

[MysteryMiner]

You are not missing something. But if network power decreases on average it takes longer than 14 days to find 2016 blocks. There also is a small variation called luck.

[shamoons]

But it's targeted to be 10 minutes. The current adjustment algorithm (as I understand it), will almost guarantee that it will be targeted to less than 10 minutes.

[Remember remember the 5th of November]

It tries to, but as it has been pointed out. Luck is the variation. There's just no real way to control block finding so it's exactly 10 minutes. It could be more, it could be less.

[MysteryMiner]

But it's targeted to be 10 minutes. The current adjustment algorithm (as I understand it), will almost guarantee that it will be targeted to less than 10 minutes.

It "guarantees" it because the network power is increasing all the time. If network total power decreases all the time, it will "almost guarantee" that it will take more than 10 minutes on average.

[Buffer Overflow]

The 10 minute block is only a guideline. It gives the system a ball-park figure to try to aim for in the next 2016 blocks when the difficulty adjusts.

[Pieter Wuille]

If the network hashing power is constantly increasing at 10% per 2016 blocks, those blocks will on average take 10% less time than 10 minutes, in the long run.

If this is what you're saying, yes.

[Stephen Gornick]

If the network hashing power is constantly increasing at 10% per 2016 blocks, those blocks will on average take 10% less time than 10 minutes, in the long run.

If this is what you're saying, yes.

To quantify that.

At the end of October 5, 2011 (UTC), the block height was 148,231 blocks.
At the end of October 5, 2012 (UTC), the block height was 201,987 blocks.

So in the past year, there was 53,756 blocks.
This was a leap year so there were 366 days.
So block generation was at the rate of 146.87 blocks per day.
There are 1,440 minutes in a day.

So each block took, on average, 9.8 minutes over the past year.

This is the hash rate (green) and difficulty levels (red) over the past year [edit: eleven months].



 - http://bitcoin.sipa.be

As you can see, even with the difficulty rate rising more than 100% over the past year, it wasn't enough to have a huge effect and blocks were still yielded pretty close to the ten minutes per-block target.

[Update:
Here's some raw data from the difficulty adjustment periods with a column seconds populated (2011 and 2012) so the computation can be made to tell how many minutes each block took for each difficulty level.
 - https://docs.google.com/spreadsheet/ccc?key=0AmcTCtjBoRWUdHVRMHpqWUJValI1RlZiaEtCT1RrQmc ]