Re: [rtcweb] Let's define the purpose of WebRTC

[Harald Alvestrand <[e-mail]>]

On 11/07/2011 02:50 PM, Eric Rescorla wrote:
> On Sun, Nov 6, 2011 at 7:20 PM, Hadriel Kaplan<[e-mail]>  wrote:
>>>> IMHO, if a web service doesn't want to take, or cannot take, the hit
>>>> for SRTP, WebRTC is not the appropriate solution for such a service.
>>> I'm exceedingly unsympathetic to the claim that SRTP is too slow. This
>>> is precisely the claim that was made about TLS for years, but measurements
>>> (see Langley and Modadugu's Overclocking SSL talk at Velocity) show
>>> that that's not really true.
>> Who said "too slow"?  There *is* an extra round-trip or two involved I presume, if we're talking DTLS-SRTP, but no I didn't mean that as a "hit".  I just meant the extra computing cycles for SRTP being a "hit".  For WebRTC-to-WebRTC I don't think that matters at all.  For WebRTC-to-media-server it might, for a free game app or greeting card app that don't care about it to begin with, and which use plaintext HTTP to begin with.
> Sorry, I didn't mean to put words in your mouth. Performance measurements
> of HTTP versus HTTPS in modern Web environments suggest that the additional
> load for HTTPS is not significant. Do you have evidence that the situation is
> different for SRTP versus RTP?
FWIW, one particularly clumsy implementation of a test of SRTP 
encryption speed that I dealt with spent nearly forty microseconds of 
(single-thread, no-hardware-acceleration) CPU time AES-encrypting an RTP 
packet - consuming nearly 0.1% of the time I had before the next packet 
arrived on that audio channel.

This, like all measurements, is false. But it convinced me that this was 
not the part of the system I needed to optimize.