This (lack of a continuation bit) would make it impossible to communicate large headers, ones which are successfully transmitted today. The alternative would be to have significantly larger frame sizes which is in no way better from a memory size or buffer management perspective. With the proposed default of 4k (which endpoints can extend to larger sized after one RTT), it is likely that we will need to use the continuation bit often enough that it should not suffer from the backup generator problem. -=R On May 10, 2013 12:28 PM, "Poul-Henning Kamp" <phk@phk.freebsd.dk> wrote: > In message < > CABkgnnWpKqAzQj1NMAF1rWKuzoErHiPKenJUgBvNzpPCx4xfeA@mail.gmail.com> > , Martin Thomson writes: > >On 10 May 2013 10:58, Hasan Khalil <hkhalil@google.com> wrote: > > >Yeah, but perhaps you can reserve that hatred for the requests (and > >responses) that have so many headers that they are larger than 64K. > >Compressed. :) > > I think we should not have a continuation bit: It will be seldomly > used, goes directly to the heart of memory management, it will > practically be an invitation for DoS attempts. > > -- > Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 > phk@FreeBSD.ORG | TCP/IP since RFC 956 > FreeBSD committer | BSD since 4.3-tahoe > Never attribute to malice what can adequately be explained by incompetence. >
Received on Friday, 10 May 2013 20:16:16 UTC