I was hoping they'd give some examples with the heap size needed for a given dataset size, but I never found them (looked through http://hollow.how http://hollo

I was hoping they'd give some examples with the heap size needed for a given dataset size, but I never found them (looked through http://hollow.how http://hollow.how also). I also wondered what you do when the heap cost gets too large. The Tooling section has a section on splitting/sharding I think answers my question. Finally, I wonder if/why keeping all the local data in RAM is a hard requirement. For example, I suspect you could build a similar producer/consumer delta/snapshot mechanism, and shove all the data into something like a local SQLite db. Or even just serialize their heap format to disk, and then mmap it. I suspect the answer is latency, and that they have the $$$ to spend on RAM and machines for sharding.