Re: GSoc2010 project suggestion: swcryptX

   Somewhat OT, but...

On Feb 22, 2010, at 19:51, Greywolf wrote:
> In such an implementation, have you considered having the ability to
> exclusively bind the entirety of a crypto handler to one cpu out of 4
> or whatever, or would that be more of a loss?
>
> [I don't see a lot of gain from 3 cpus up to 4, but then I'm not
> benchmarking, either.  If I were doing accelerated crypto, I
> would consider it acceptable to have a single cpu of the multi-
> core system dedicated exclusively to handling crypto.]

   I've had some similar thoughts.  Not doing a lot of benchmarking,  
but using lightly-loadd real-world workloads, I see a *clear*  
advantage in 2 processors.  Both desktop and server benefit.  But,  
I've been wondering of the advantages of four.  I mean, certainly, the  
"right software" makes great use of such, but few are the "right  
softwares" in my experience, and they aren't the general-purpose  
systems.  (more now than 10 years ago, but....)

   I find interesting the idea of dedicating a CPU to some tasks like  
crypto, that are CPU expensive.  If the kernel and/or libs were  
engineered such that any vaguely multi-thread-capable application need  
do nothing to have a "marked" CPU handle all the crypto, and the other  
threads handle everything else, it does seem kinda cool...

   And of course, the same invitation for dismissium applies...

              - Chris