iperf3 is a real Internet application in the same way that ping, traceroute,
tcpdump, etc. are real applications. It is a well-known tool that network
engineers use on a daily basis and demonstrates that UDP performance
is highly correlated with UDP datagram size (i.e., even for sizes that
exceed the PMTU).

> -----Original Message-----
> From: Fred Baker [mailto:[email protected]]
> Sent: Thursday, November 29, 2018 12:53 PM
> To: Templin (US), Fred L <[email protected]>
> Cc: Ron Bonica <[email protected]>; Joe Touch <[email protected]>; 
> Stewart Bryant <[email protected]>; int-area
> <[email protected]>
> Subject: Re: [Int-area] draft-ietf-intarea-frag-fragile-03
> 
> 
> 
> > On Nov 29, 2018, at 11:02 AM, Templin (US), Fred L 
> > <[email protected]> wrote:
> >
> > If you are referring to iperf3, that is exactly what it does (uses 
> > fragmentation
> > to generate a high data rate), but I would not call it "artificial" - 
> > fragmentation
> > is a real Internet service.
> 
> But iperf3 is a test/diagnostic tool, not a "real" internet application. The 
> reason for bringing up NFS is that it was, at least at one time, a
> real service that was widely used in real networks, and breaking it would 
> have been a real operational problem. Even for that, since
> fragmentation would likely occur at the originating host (as in IPv6), apart 
> from tunnels or links that reduced the PMTU, the issue
> would have been between the endpoints and any middleware en route, and since 
> NFS didn't cross network boundaries, it would be
> unusual to find firewalls in the path.
> 
> For NFSv4, RFC 5661 requires the use of TCP, SCTP, or "whatever is 
> recommended at the time", which could perhaps refer to QUIC
> today, and discourages the use of UDP. It also refers to RDMA etc.
> 
> I frankly don't see what "observing the demonstrable and measurable fact that 
> there are issues with Internet Fragmentation and
> Reassembly" hurts. The paper doesn't recommend that the service be deprecated 
> or removed. It does, however, suggest that
> applications and transports not depend on Internet fragmentation and 
> reassembly; applications above the IP layer that need packets
> smaller than the relevant PMTU should find a way (such as by using TCP) to 
> avoid it. I don't see how iperf3, or for that matter any
> application, is harmed by that.
> 
> I think that what we really need to understand is what measurable harm this 
> comment and recommendation causes. Maybe you
> could start your reply with that issue?
> --------------------------------------------------------------------------------
> The fact that there is a highway to hell and a stairway to heaven is an 
> interesting comment on projected traffic volume...

_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area

Reply via email to