Any benefits to use gRPC over QUIC?

[Jeffwan]

Hi @gfanton Thanks for open source this project.

I am new to QUIC and investigating whether QUIC is beneficial to gRPC. Could you also share some insights on the benefits to use gRPC over QUIC? Did you run any benchmark on QUIC vs HTTP2? I am trying to understand if the problem statement makes sense.

Thank you!

[gfanton]

Could you also share some insights on the benefits to use gRPC over QUIC

I think the only advantage at the moment is to use it on a mobile.
By using QUIC+UDP, we will have a packet-oriented connection, so you shouldn't have any loss when switching from one network to another (4g/wifi/...), It Also should works better than TCP in a flaky/slow network.
Overall, QUIC makes all your connections more reliable, which makes more sense on a mobile or a browser.
Otherwise, I don't see any other significant advantages for now (but maybe I missed something).

Did you run any benchmark on QUIC vs HTTP2

I may be wrong, but until GRPC natively supports QUIC, what it actually does is HTTP/2 over HTTP/3 (QUIC), so I don't think there will be any improvement in performance, or maybe some latency, but I don't think it will compensate the overhead. Perhaps GRPC over KCP-GO (which has no built-in muxer for example) should be a better choice than QUIC.
If you are still interested, you are welcome to add some benchmarks.

[Jeffwan]

@gfanton Thanks a lot for these details. In this case, if the gateway layer support QUIC (server side) and mobile supports QUIC (client side), that should be good enough. After gateway route the traffic to internal microservices (managed by service mesh). Since the network latency and stability are very promising, QUIC won't help a lot in microservice inter-communication? I will definitely look at KCP-GO. thanks for the pointer