Re: #540: "jumbo" frames

On Wed, Jun 25, 2014 at 05:28:47PM -0700, Martin Thomson wrote:
> Sure, we might have arrived at what is only a local minimum, but
> without stronger justification I'm really reluctant to act on this.
> As far as it goes, Willy's numbers don't actually concern me that
> much; parallelism goes a long way to addressing those sorts of
> concerns.

Martin, while I can understand that such numbers are irrelevant to
your use case, and that you're not tempted by a last-minute change,
I'd like to mention that parallelism is orthogonal to this concern ;
parallelism is what currently makes it possible to reach close to
100G with HTTP/1.1 and if the same hardware goes back to 10 or 20G,
it's not parallelism that will bring the performance back, it's just
a definitely wasted performance by a design which does not scale as
well as the one it replaces.

Regards,
Willy

Received on Thursday, 26 June 2014 05:53:48 UTC