Bitcoin Forum

I've been watching the total hashing speed daily and i've noticed it fluctuating between 8.9-14.7 th/s, and looking on the forums for an answer to what's causing it. Does anyone have any ideas other than botnets? Almost a 40% drop seems a bit much to chalk up to pools and/or personal miners losing connectivity.

I think it's important to remember that the hash rate is an estimate based on production rate and there's randomness in rate of production.  Sometimes the network is really lucky and produces several blocks in a very short span, even as quickly as seconds between blocks.  Sometimes the network is really unlucky and it takes a lot longer than 10 minutes to produce a new block.  Those variations are reflected in the estimates.  That, of course, isn't all there is to it, but it does play a noticeable role.

hrm, I had forgotten about that. I suppose that does explain some of the fluctuations, but it doesn't answer the long term rise and fall. i.e. when we jumped from 1.5 to 1.79 and then back down to 1.59.

Well 1.5 to 1.79 is <20%.  Prices rise, miners see profits and turn on rigs (that may include bonets who allocate nodes from other tasks to mining).  Difficulty rises, miners quit, difficulty falls.

The revenue per GH/s has been pretty stable for some time now.

http://bitcoinx.com/charts/chart_large_lin_90d.png

1 GH/s earns roughly $3 to $4 per day.  When prices rise it pulls in marginal miners (april to early may) which causes network to solve blocks faster than 6 per hour.  When the resulting difficulty rises it makes marginal miners unprofitable and they quick.  This is reflected in later difficulty adjustments.  There is some lag and plenty of noise but revenue per Gigahash is remarkably stable.

I wonder if summer has something to do with the difficulty drops. Some miners may not be able to pay for cooling those rigs?

quick tip, Bitcoin is worldwide
I have overclocked my cards more - it is winter here  ;D

This accounts for ~90% of the fluctuation. If you think about it, in the WHOLE world, only one BitCoin block is solved every 10 minutes. Sometimes, the block takes 15 minutes, and others, 5 minutes. This where the wild fluctuation occur, and why you need to look at a longer data set to get a more general look at the network.

Either that or it's BFL cycling their farm of MiniRigs before they ship  ;)

This doesn't even describe it enough. Sometimes its <1 minute between blocks, hell sometimes its even an hour+.

It seems like it should be possible to calculate the expected variance in the hash rate estimate on a given window size.  

For example, the 8 hour window estimate is based upon the time required to solve 48 blocks.  You should be able to calculate a 95% confidence window around each 8 hour estimate.  I'm not in a mood to crack a stats textbook tonight, but I am curious about the subject.  A good analysis might lead to better ways to smooth the estimates than simple windowing.

By eyeball there seems to be some statistically significant periodicity to the data sets.  In particular, there seem to be large upswings on the weekends.  Since it makes no sense to not run hardware purpose built for mining mostly 24x7, I suspect this is the result of corporate or university workstations being farmed by admins.  It could also be someone taking advantage of lower priced power and running otherwise unprofitable systems.

Would like to see distribution of miners though. If a large amount of miners are in the US, they would overclock just as much as you in the winter, and since there's more of them, there would be a net network hash-rate increase in the summer and a decrease in the winter.