[NNTP] Re: Comments on draft-ietf-nntp-tls-nntp-05.txt

[EKR]

"Forrest J. Cavalier III" <forrest at mibsoftware.com> writes:
> Andrew Gierth wrote:
>
>>>>>>>"EKR" == EKR  <ekr at rtfm.com> writes:
>>  >> Well, nondisclosure limits how much I can say, but it's
>>  >> unquestionably true that (a) traffic levels of many gigabits are
>>  >> the norm rather than the exception in the commercial Usenet
>>  >> provider industry (which is a very significant user of
>>  >> authenticated NNTP connections, and more importantly also has a
>>  >> major effect on client development) and (b) the CPU cost of
>>  >> encrypting all that, purely to protect the password, is not
>>  >> something that can simply be absorbed.
>>  EKR> Yeah, this falls more into the category of assertion than
>> data. What sort of data would you like?
>>
>
> Does Andrew need to disclose CPU loads to make this point?

Well, some kind of measurement, yes. 


> A rough estimate of throughput on a GigE is 100MB/second. At that
> rate, with a 3.2 GHz processor clock, you get 32 clock cycles per byte
> on average.  Those 32 clock cycles are precious:  to the
> extent that you don't fill the I/O pipe, you need/waste hardware
> resources (network and CPU.)
>
> The real cost of encrypting is higher because the memory bus
> runs at a slower speed.

Yes, I'm familiar with all this, but what's relevant here is not
the absolute cost but the relative cost compared to other things that
are taking up CPU (Amdahl's law again). That's why you need
actual measurements.

-Ekr