Skullmfer🎩

$kas, I am excited to announce the first episode of the Kaspa Deepdive! And just in the nick of time too, with just over 21 hours to activation! To sweeten your agonizing wait, I invite you on a leisurely tour of what makes high BPS awesome. Join me for the Boons of High BPS.

$kas, I don't deserve the credit I get for Crescendo. Posts that credit me for Crescendo make me uncomfortable, especially as they rarely tag people who I know contributed much more than I have. I made a handful of conceptual contributions, of which I am proud, but measured against the total scope of work behind RK and Crescendo, they are minuscule. So far, I have chosen to ignore such posts categorically. But as the activation looms, and they become more frequent, I realize it is important to me to be more explicit. My place in the Crescendo festivities is celebrating, not celebrated. The people you should celebrate are @michaelsuttonil, @OriNewman, @coderofstuff_, @Tiram_88, @Max143672, and quite a few others (less time I made such a list I was scolded for including someone irrelevant, so I'd rather err on the side of omission, feel free to fill in the blanks in the comments).

A good mental exercise for making good decisions is trying to devise bad decisions that might seem good on the surface. So I sometimes ask myself, what is the simplest, most straightforward modification I could do to $kas that will seem OK but will actually destroy its security? I think an obvious answer is messing with the finality depth. The finality depth is a parameter in the system that essentially means "if we ever have a fork that goes deeper than this depth (on both sides), we need to halt all operations and address the situation via a community discussion". Once this parameter is set, we can implement pruning. For reasons I won't get into, the pruning depth is slightly more than twice the finality depth. So it is tempting to reduce the finality length to save storage. If you make it ten times shorter, you only need to store ten times less ledger data. So what's the problem? Well, as I said, once you allow pruning of this depth, you can't automatically recover from a split longer than the finality depth, as you don't have the required ledger data to verify both sides of the split. A situation where an adversary is trying to cause such a split is called a pruning attack, and setting the finality depth is essentially setting the price for a successful pruning attack. So what does that mean in concrete numbers? Well, say, for the sake of discussion, that I have a Kaspa fork whose entire hashrate can be produced by about one thousand 3090RTX GPUs, and I decided to reduce the finality depth to 30 minutes. What does that entail? Renting a single 3090RTX by the minute can be as cheap as 5 cents an hour, but you would probably not be able to procure a thousand GPUs for the cheapest price. However, it is completely reasonable to rent one thousand 3090RTXs (or equivalent) for 10 cents an hour each. Since I only need to attack the network for half an hour, that's 5 cents per card, or about $500. But what if I already have a sufficiently large mining rig? What would that cost? In that case, the cost of attack splits into two: the power expenditure, and lost profits from diverting the GPUs. According to energysage.com, the average power cost is 19c/kWh. The maximal power consumption of a 3090RTX is 350W/h, so running a thousand of them for half an hour requires about 175kWg, or about $14.25. According to whattomine.com, the most profitable coin to mine on an 3090RTX at the time of mining is something called Flux, with a revenue of $0.54 per 24 hours, or about half a cent per half hour. Multiplying by one thousand cards, we get $5. So for a miner with access to this amount of cards, a successful pruning attack that will completely crash the network would cost less than $20. If a thousand cards sound like a lot remember: a. this attack could be shared between owned and rented GPUs, so if you have less you can rent the rest, and b. in 2022 alone an estimated 14.2 million of Nvidia's GPUs were purchased for mining. For comparison, in Kaspa, the finality depth was set to 24 hours at launch time and has never changed since. The Crescendo update will reduce it to 6 hours, which is perfectly secure now that the mining market is capEx heavy, ASIC dominant, and worth hundreds of millions. This is a great example of how a seemingly innocuous change can tear through the network security, and why, if you choose to fork a tech you do not fully understand, you should be very careful when tinkering with its innards.

Hey $kas. I made an animated demonstration of why ten times the BPS does not imply ten times faster confirmation times. Essentially an animated version of my post from a few weeks back (link in comments)