W3C home > Mailing lists > Public > ietf-http-wg@w3.org > October to December 2013

Re: TLS at transport level vs stream multiplexing and aggregation (http "routers")

From: Willy Tarreau <w@1wt.eu>
Date: Mon, 18 Nov 2013 00:30:28 +0100
To: Roberto Peon <grmocg@gmail.com>
Cc: Nicolas Mailhot <nicolas.mailhot@laposte.net>, Adrien de Croy <adrien@qbik.com>, Mike Belshe <mike@belshe.com>, "ietf-http-wg@w3.org" <ietf-http-wg@w3.org>
Message-ID: <20131117233028.GF18577@1wt.eu>
On Sun, Nov 17, 2013 at 03:06:49PM -0800, Roberto Peon wrote:
> There are interesting security implications of interlacing unencrypted and
> encrypted data which I'm fairly sure have not at all been
> analyzed/experimented with.
> This was one of the reasons why we originally thought about, but did not
> implement, encryption as an upper, rather than lower layer.

I remember another point which was about the number of TLS contexts over
a single connection which would make the memory usage explode. This is
still true. With cryptography, it's not possible to have an "Y" shaped
device which saves both CPU and memory at the same time :
  - either you want to save decrypt/encrypt cycles and you route encrypted
    messages with their own contexts, buffers, etc. Here however for routing,
    mostly the endpoint (origin server) has to deal with that many contexts.

  - or you want have only one context and have to encrypt/decrypt.

> WS does do masking, but that is a fair bit less involved than TLS, which
> requires bidirectional communication and is more involved.
> 
> Again, I think it is an interesting thing to experiment with, and think
> that it will absolutely require lots of analysis and experience...

Yes, as usual. I just hope we won't experience that one-year dance we had
with WS...

Willy
Received on Sunday, 17 November 2013 23:30:55 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 1 March 2016 11:11:19 UTC