Jeff Dean's talk about rapid response times

[mrjn (Manish R Jain)]

Update: We ended up switching to Protocol Buffers, after Flatbuffers proved to be slower. Also, Flatbuffers API is complicated and ugly; so once the benchmarks proved that FB is slower for our data, it was an easy decision to move to PB.

Comparison of Protocol Buffers v/s Flatbuffers

I ran benchmarks to marshal, then unmarshal a list of uint64 uids via Gogo Protocol Buffers and Flatbuffers. Here’re the results. $ go test --bench . --benchmem ~/go/src/github.com/dgraph-io/experiments/flats testing: warning: no tests to run BenchmarkToAndFrom/Proto-10-4 3000000 556 ns/op 440 B/op 7 allocs/op BenchmarkToAndFrom/Flatb-10-4 2000000 717 ns/op 440 B/op 12 allocs/op BenchmarkToAndFrom/Pr…