Joe Hellerstein
Joe Hellerstein
these are ostensibly separated because there's much more serving traffic than routing traffic, and we can deploy many fewer routing nodes. Still, doesn't seem very harmful to have that lightly-used...
We should have a benchmark directory and keep benchmark code under git. Would be nice to have performance regression suites if feasible.
- Can we do something native in C++ that's extremely readable? - And/or do we want a portable, language-independent syntax for dataflow specification that is easily parsed?
Needed to reason about "before" and "after" non-monotone operations
- C++ client library - Documented API with basic parameters - startup configuration (ports etc) - command set and parameters - A REST wrapper for server so clients can be...
We want to "whitelist" monotonicity, and carry the "taint" of non-monotonicity through the type system. Possible approaches: use C++ traits