Re: [hybi] Masking overhead with measurements and code

[Greg Wilkins <[e-mail]>]

On 9 January 2011 16:14, Eric Rescorla <[e-mail]> wrote:
> On Sun, Jan 9, 2011 at 1:43 AM, Greg Wilkins <[e-mail]> wrote:
>>
>
> Wow, I think you have this completely backwards: it's in systems that
> do almost nothing
> that the cost of adding a new, expensive operation is large. In
> systems that are already
> doing a lot of other processing, the marginal cost is comparaitvely
> low even though it may
> occasionally push you over some threshold; Amdahl's law and all that.


Eric,

I did not say system that are "already doing a lot of other
processing", I said systems where parsing is around 5% of the CPU
load.   This may be for a high message rate on a large server or it
may be for a low message rate on a constrained device.   For such
systems, multiplying the CPU work needed to parse a message several
times will have significant impacts.

Besides, it is self evident that adding cryptographic processing to a
protocol puts additional loads on application servers - we need only
look at SSL and the existence of SSL offload devices.

regards