[e2e] Is a non-TCP solution dead?

Mark Allman mallman at grc.nasa.gov
Tue Apr 29 11:48:21 PDT 2003


> or middleman - think 'performance-enhancing proxy' doing ack
> splitting/doubling (though RFC3135 section 3.1 seems a bit coy on
> this method.)

Right.

("We're the world's fastest ISP!")

> In preventing this, ABC is obviously sane - and sanity in the face
> of a savage attack is a good thing. This is why ABC should go
> standards-track.

I agree -- but, I think it is a bit more dicey than that.  If the
rule were simply "increase by the number of bytes ACKed, but no more
than an MSS" then that would be fine.  And, that would have gone
standards track easily, I think.

But, since there is some language that allows you to increase by the
number of bytes ACKed up to 2*MSS it is more aggressive than current
TCP and so in need of a bit of caution.

My bet is that if we were to revise rfc2581 (which seems like maybe
needs to happen) we'd include the variant of byte counting where you
can only be more conservative, not more aggressive.

> more experimentation? It was a draft for three years.

Well, ... blush...  *In this case,* that delay can be *at least*
equally attributed to me scraping cycles and not to the IESGs
slowness.

allamn





More information about the end2end-interest mailing list