Birth of the NearCloud: Serverless and CRDTs Edge Is the New Next Thing
highscalability.comVery informative! Made a nice Monday read. I specially liked the analogy with the Amazon Fulfillment centers. Pretty amazing to see that there is so much untapped potential for latency improvement and Kuhiro taking the steps in just the right direction. Would love to read more about CRDTs and the internals of this platform. Good luck!
They mention Google Cloud Spanner as being slow due to locking during the two-phase commit. However would distributed transactions be possible without locks?
If I’m not mistaken, CRDTs cannot provide distributed transactions.
Two phase commit (in spanner) is similar to a lock, so the short answer is YES you need locks for distributed TRANSACTIONS
But you dont need transactions to do distributed computation. This is where CRDTs come in, they allow you to modify data locally w/ zero wait and the replication of these modifications happens asynchronously (imagine a split second later) and then there are algorithms that run on the receiver of the replication that automatically and intuitively resolve any conflicts that happened because the data was not (globally) locked
I will have a post next week on serverless.com explaining the internals of CRDTs, it should help clear stuff like this up.