TLDR, split up a blockchain into a bunch of smaller blockchains sharded by something (let's say address), and then a different tier of nodes which only operate at the shard level can exist -- they call these "collations". collation-internal transactions can be managed by only the nodes in the collation, and they even lay out an initial proposal for how to do cross-collation transactions, leaning on merkle proofs for verification. They also propose making it transparent as a part of an auto-loadbalancing part of the blockchain so developers wouldn't feel any difference.
This paper was actually pretty interesting. It's pretty thoroughly researched, and the solution that they come up with (sharding) is actually pretty promising as a solution to the centralization + throughput issues of the blockchains up until now. I certainly learned a lot about blockchain systems by reading it, and more importantly learned the current problems with the state of the art and how this group plans to address them.
At this point, I also look at blockchain as a generic distributed database problem -- so I get the feeling these finndings will benefit the field some day. Surely something will come out of the research energy that's being dedicated to blockchain right now that will be beneficial to future researchers.
The details haven't been decided yet, but developers will need to deal with some kind of locking or "yanking" [1] mechanism in order to communicate with accounts on foreign shards.
TLDR, split up a blockchain into a bunch of smaller blockchains sharded by something (let's say address), and then a different tier of nodes which only operate at the shard level can exist -- they call these "collations". collation-internal transactions can be managed by only the nodes in the collation, and they even lay out an initial proposal for how to do cross-collation transactions, leaning on merkle proofs for verification. They also propose making it transparent as a part of an auto-loadbalancing part of the blockchain so developers wouldn't feel any difference.
This paper was actually pretty interesting. It's pretty thoroughly researched, and the solution that they come up with (sharding) is actually pretty promising as a solution to the centralization + throughput issues of the blockchains up until now. I certainly learned a lot about blockchain systems by reading it, and more importantly learned the current problems with the state of the art and how this group plans to address them.
At this point, I also look at blockchain as a generic distributed database problem -- so I get the feeling these finndings will benefit the field some day. Surely something will come out of the research energy that's being dedicated to blockchain right now that will be beneficial to future researchers.