[DennisP]

Of course it's not optimal. But consensus mechanisms resilient to Byzantine attack are an active area of academic research, and traditionally they've all required all nodes to hold the complete state. Bitcoin's contribution was a new method that scaled to a much large number of nodes and didn't require a known set of nodes; the tradeoffs were low throughput, long latency, and offering only probabilistic finality.

But it wasn't Vitalik who created that; what he did was extend it into an application platform. That platform is working, with live applications including decentralized exchanges, Maker (stable currency and collateralized loans) and Augur (prediction market with decentralized bet resolution).

Sharding is one of the main things he and the research team are working on now (along with proof of stake). The current design uses 4000 shards, so unless you consider 1/4000 to be a large subset, it would answer your objection if they manage to get it working.

[lavrov]

Verifiable computation is an active and developed area of research in cryptography, and the validation method of having every node run every step of every computation is the most naive approach. There are plenty of ways to prove to you that I faithfully executed a computation that don't rely on your also having performed the computation.

I don't know how active the Maker userbase is, but Augur has 700 monthly active users.

It seems like it would be possible to manipulate the validator set for a particular shard - also, isn't part of the point of ethereum to be able to access information from other contracts? How do you guarantee that that contract is on the same shard?

[DennisP]

People are working on things like Truebit and zksnarks for Ethereum, but those still depend on base layer consensus for posting the proofs. They just reduce on-chain computation.

Augur just launched a couple weeks ago.

Preventing validator set manipulation is an important part of sharding research. Part of the solution is getting good random numbers that can't be manipulated, and there are several approaches for that.

Cross-shard communication gets complicated. There are some ideas for it, but that's several years down the road. It'll still be a big step up to have lots of shards, each with the capacity of the full blockchain today, sharing the security of all of them but not getting swamped with traffic if something gets really popular on another shard.

[wycs]

There is very active work on using STARKS for this purpose. In the long term this is the plan for both Tezos and Ethereum.

[daics]

> I don't know how active the Maker userbase is

$1M in new loans issued every 3.5 days for the last 7 months at 0.5% APR.