Http2 slower then http1 in high volume

[ricardkollcaku]

Im trying to do some benchmarking with spring boot using http1 and http2
and i see i weird behaviour.

Im using for client and server
SPRING BOOT 2.6.6
Java 17
Both using webflux

http2 is bit slower than http1 looks like is not taking advantage of single tcp connection with multiplexing.

you can find the project in https://github.com/ricardkollcaku/http2-test

My Server:
I have enabled http2 in properties :
server.http2.enabled= true
and in using curl

curl -I --http2 http://localhost:8080/sms-request/test
i get:

HTTP/1.1 101 Switching Protocols
connection: upgrade
upgrade: h2c

HTTP/2 200 
content-type: text/plain;charset=UTF-8
content-length: 10

My client is another spring boot 2.6.6
and looks like

 WebClient webClient = WebClient.create().mutate()
       .clientConnector(
            new ReactorClientHttpConnector(
                HttpClient.create().protocol(HttpProtocol.H2C)))
        .baseUrl("http://localhost:8080/sms-request/test").build();

    Flux.range(1, 100000000)
        .concatMap(integer -> webClient.get().retrieve().bodyToMono(String.class))
      .subscribe()
    ;

Using default protocol i can process up to 23,959 requests per second
with H2C protocol i can barely reach 16.000request per second

[violetagg]

@ricardkollcaku We are working on fixes related to the observed behaviour.

[violetagg]

@ricardkollcaku

In the reproducible example you are using flatMap

    Flux.range(1, 100000000)
        .flatMap(integer -> webClient.get().retrieve().bodyToMono(String.class))
        .subscribe()
    ;

While in the comment above you are specifying concatMap

    Flux.range(1, 100000000)
        .concatMap(integer -> webClient.get().retrieve().bodyToMono(String.class))
      .subscribe()
    ;

Please clarify the scenario!

Also with curl you are using connection: upgrade while the WebClient is configured to use H2C prior-knowledge. Please clarify the scenario that you are interested in. Here you can find more about WebClient configuration related to these two scenarios: https://projectreactor.io/docs/netty/release/reference/index.html#_protocol_selection_2

[ricardkollcaku]

Hi @violetagg
The example was just for test (benchmark) propose and i was trying to check how http1, http2 and rsocket compare with each other in terms of speed and i was expecting that with http2 to be faster than http1 but in my benchmark http2 is performing slower than http1.

Yes i have tested it with flatmap and concatmap. i thought maybe there was not working multiplexing with flatmap thats why i tested with concatmap.

But the problem is not flatmap or concatmap is why http2 is performing worst than http1