I agree that this limits the practical damage in this specific scenario, but for the sake of argument I’m trying to imagine that this scheme is being used for other things too.
Sure, let’s imagine that I think the main limitation is that such a system without full collateralization cannot be used for buying stuff that you can “consume” quickly or transfer/resell. It may be similarly suitable for incentivization of other protocols too. For higher level applications the “lightning style” multi-hop approach sounds better, especially because the probabilistic behavior is also a serious inconvenience when paying higher amounts.
Also, the other side of the question: What is a minimum fee required to sufficiently encourage stampers to exist? I suspect the floor will ultimately set by whatever fees you earn as a Casper staker, as a percentage of your deposit.
We should also write up some back-of-the-envelope estimates for potential host earnings in different scenarios (e.g. say we convert 10% of the network into paying light clients). This will inform what kinds of deposits we could expect from stampers.
First I’ve made some very rough estimates for LES service assuming a big enough network with enough players so that market mechanics can work:
- a home PC can serve about 50-100 clients with minimal capacity (equivalent to free service)
- earning 50-100 USD per month should be enough incentive for many average users to run a light server (almost) 24/7 so the price of a “minimal capacity slot” (MCS) should not be higher than cca 1 USD per month
- having 5 MCSs at different servers ensures a high-quality, high-reliablility connection to the network (for most use cases 2 might be enough)
- 5 USD per month equals 0.7 cents per hour. For a regular user who is not constantly connected this service is definitely cheap enough.
- for 24/7 connected applications the 2-5 USD/month price may still be suitable or alternatively they can go for their own full node. In the future the pricing models can become more sophisticated (giving a discount to those who are really just syncing headers 99% of the time) but we can already say that even with a really simple pricing model there will be demand at 1 USD/MCS/month.
For the price of the stamper service it’s really hard to make a guess at this point but I think there is a good reason to assume that it will work itself out. The main reason for a stamper to exist is that paid servers require them in order to be able to accept payments so stamper service will probably be provided by some of the servers themselves. I would go as far as making the stamper protocol an optional part of the LES server side protocol. Since stamping only requires receiving, signing and sending few hudred bytes packages cca every minute per client-server connection, a PC capable of running LES service can very easily also handle the stamping needs of 10 or even 100 other servers. If there is a stamper for every 10 servers and we accept spending 1% of the revenues on stamping service then the estimated extra revenue one can gain by having a deposit and allowing stamping is 10% of the service revenue or 5-10 USD per month. If we demand 100 USD deposit this sounds perfectly fine but if there will ever be a shortage of stampers at this price, the server to stamper ratio can easily go up to 100 while the server revenue to stamper cost ratio can also raise to a few percents, allowing hudreds of dollars of stamper revenue if necessary (which I really doubt).
Of course starting off from the current state of a non-existent market is a different story but right now a few stampers can easily serve the whole world and I would definitely run one just to be safe, regardless of the revenue Also I am not sure about LES service prices at the early stages but it even kind of works almost good enough without any payment and most of the server operators will turn on accepting payments because why not. Some clients will also try paying when they get shitty service for free (which happens a lot) so I guess it will figure itself out somehow.