Thursday, November 3, 2016

Getting the rough queueing numbers for the pit

Use some very approximate assumptions.  Take shannons equation,
1+ SNR.  SNR is about 6, the amount  of signal power to noise, noise in this case being residual disorder on a partially complete tree.  But, a unit of power is a step on the graph, as nearr as I can tell.  So we know right away that trading bots, signals,will be about 65% ofthe graph cycles, 7/8.

But makingn the tree rrquires NlogN, N being the depth, which will go roughly as SNR.  Bot search time goes as N.  Hence, to a good approximation, he pit can maintain about N^2 bots queued on the tree, for an SNR near N.  But the bots are moving on and off after N steps, and the pit boss is constantly flushing in new bids

No comments: