Sign into an existing account

You mean bigtable? Never used it, but tried most amazon equivalents (dynamo, redshift, athena, documetdb, s3) as well as cassandra+spark. I thought cassandra was really annoying to optimize queries for. I liked how seemless documentdb was to use with json payloads. Dynamo is probaby closest to bigtable and I found it easy enough to use but wish it was schemaless.

Side note: s3 is surprisingly well suited as a key value store if you front cache it.


Haven't heard of google's simulation system- got a link? I really like reinforcement learning so maybe it's a good platform to do that over?


Haven't read it but open to doing so.

Not a hardware RNG, no - if you want true hardware entropy then you need something like radioactive decay, same as with a high-precision time source. Most computers don't have that and must rely on random events occurring (such as network packets), which blocks /dev/random if they stop occurring, hence /dev/urandom for non-blocking PRNG.

The point of this development is that clock jitter always occurs over time, so adding that as an entropy source means that /dev/random is always fed, hence using that for everything. Technically it can still block, though the jitter entropy means it can no longer block forever (or more than a moment, in practice).