Ok let's pretend I know nothing about distributed systems (we won't have to pretend very hard): why is it silly to think something could be slow because it takes 1000 RPCs to render?

@ipvkyte But an RPC is different from a network hop, no? Isn't there a bunch of per-call overhead?

@moyix @ipvkyte Connections can be reused and HTTP supports pipelining, so if multiple calls are made in parallel, they can reuse the same connection at the same time. The overhead is quite minimal. The goal is to minimize dependencies between requests to maximize concurrent reqs.

@tixxit @ipvkyte I think this would potentially be covered by "poorly batched"? But I think what I'm getting from other replies is that it's not implausible, just wrong

@moyix @ipvkyte I’m replying to the “overhead of network calls” comment. But, I think the most likely interpretation of “poorly batches” is that Elon just meant the requests were small/single purposed. I do not think for a second he was referring to pipelining or concurrency.

@moyix @ipvkyte I think his idea of batching is “doing multiple things in a single http request.”

@moyix @ipvkyte So, “poorly batches” was referring to “not putting enough stuff in a single request.” That said, I don’t think it’s an unreasonable worry. I think blasting it on twitter is bad. Treating your eng like clowns instead of smart, capable people is bad.

@moyix @ipvkyte But I if a leader asked me the same question in a meeting or in private, I would 100% think it’s reasonable and would explain.

@tixxit @ipvkyte Yes, totally agree it's very poor form!