This is why I proposed that they shut down or spin off their own nodes (in an orderly manner) and only then undertake to optimize the code.
Agreed...
There is a real tradeoff with RocksDB in that it will likely be a lot less efficient in terms of performance while reducing memory usage.
RocksDB using local persistent store on every node along with fault tolerance. (change/commit log).
Isolated so any issue with the local store on a node does not affect other local stores on other nodes and the required remote store.
If the local state size is big, time to restore the state during a restart also takes time. That's the only tradeoff I can imagine. Maybe data duplication is another one as the same data is also needed on the remote store.
True parallelism of data access and data updates with fault tolerance ensured. :)
BUT!!! Is this really needed? At what costs? :)
Significant understanding, effort and set up is required to pull this off.
Did they make and discuss a cost-benefit/impact analysis with all the witnesses?
It is one company deciding that its own short term AWS bill should determine the course of the development roadmap for the entire blockchain, and that's a problem.
Agreed...
There is a real tradeoff with RocksDB in that it will likely be a lot less efficient in terms of performance while reducing memory usage. That will have a long term cost in terms of scalability of transaction rate and then ultimately other complexity and maintainability issues (which is why graphene was designed with a non-blocking in-memory model for transactions validation in the first place).
Agreed...
Memory usage is not the culprit but the snake oil, it is exactly what you say scalability aka bandwidth issues. Graphene is the blockchain transaction processor, we have no problems there, also not in the future when those millions upon millions new users are coming on board. (quoting ned, or was he dreaming about dollars).
With increasing complexity, maintainability and therefore costs are on the rise.
Did they handout a hardware requirement shortlist to the witnesses to run a witness node (incl. RocksDB local store)? Or maybe running a witness full node is only for the few? Or maybe, they are not planning to use local persistent stores? evil laugh
Its not going to happen... witnesses will not allow them to offload those AWS bills into their wallet.
Sorry to be realistic again.
Steemit inc, is not the next google, facebook or linkedin...
the only problem is you have alot of unchanging historical data that really needs centralized repositories, kind of a star network. Modern Databases are optimized for this, and no matter how much decentralized bandwidth you throw at it, the coordination and transmission speeds run by a factor of 10,000 times slower AT MINIMUM then even a crappy database on a server.
There needs to be two different tiers of nodes, daily operation blockchain is perfect for decentralized dynamic block chain, And then heavily indexed Historical Repositories on a smaller regionized distributed network.
The fact that old data is queried at a rate of 1/100 vs Active Social Media.
So say 1000 witnesses boxes, then maybe there would be 5 or 6 historical data boxes