Re: [core] CoAP for high throughput applications

[Carsten Bormann <[e-mail]>]

> A slightly off-topic question -- though not too much, hopefully.
> 
> One of the LURK [0] proposals is draft-cairns-tls-session-key-interface [1].
> 
> In Section 7.1.2 [2] the authors propose to transport the LURK payloads
> over CBOR/CoAP (over DTLS/UDP, I guess).

Interesting.

> Now, a single LURK box could have to handle lots of these requests,
> potentially in thousands per second, whereas CoAP's default congestion
> control algorithm parameters [3] are, by design, way too conservative to
> be suitable for high-throughput use cases.

That would be an interesting application for CoCoA...

> Is there anyone that has played with CoAP for high-throughput
> applications who'd be willing to share his/her experience with the group
> and the wider IETF community?

I think Matthias Kovatsch' Californium still is the record holder for
responses per second from a Web Server (a term that I take to include
both servers for CoAP and for one of the HTTPs)...

The 16-bit message ID is only good for ~ 10^3 requests per second, so a
client that needs a higher rate from a single server would need to use
multiple port numbers.

Grüße, Carsten