[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Perfomance problems and main stream




On Tue, 23 Feb 1999, VALETTE Eric wrote:

> >>>>> "Quality" == Quality Quorum <qqi at world.std.com> writes:
> 
> Quality> On Mon, 22 Feb 1999, Mark Johannes wrote:
> 
> mark> I agree that the BSP is the area of most diversity, hence the problem. 
> mark> What we are really attempting in to provide a framework and architecture 
> mark> that allows this diversity to be applied without hindrance.  Therefore, I 
> mark> believe this will assist the newbie when a clear architecture is presented 
> mark> for the structure that he/she has to integrate the application hardware 
> mark> within.
> 
> And I follow mark. From experience, the key success for a RTOS is the
> number of drivers it supports. The grow in this area is exponential
> as, the more driver you have the less code you need to add to provide
> a new one. BSP is thus a mid term key success factor.
> 
> Quality> 1. Is too application specific, so I doubt that 
> Quality>    it will be ever possible to coin it into something as universal 
> Quality>    as BSP.
> 

Let consider simple router application as an example.

> How mange Linux to work on MIPS, ALPHA, INTEL, PPC, ARM on various platform?

Can Linux perform routing ? Yes.
Is it competitive vs specialized routers ? No

> Same for VxWorks?

Can stock VxWorks IP stack and drivers perform routing ? Yes
Can you build good router this way ? No

> 
> I'm really tired to write drivers nearly from scratch for new boards and
> having to understand all the complexity of processor/system dependent
> interrupt mechanism, cache species, DMA handling...

Can RTEMS IP stack route ? Yes
Can you build good router this way ? No
Do we have any chance of substantial improvement in this area ? No, and 
there is no reason even to waste time on it.

> All this can/should be hidden by PROCESSOR API/BSP API.

IMHO, it is simply impossible requirement. IMHO, the goal of BSP 
should be to have working set of routines suitable as starting 
point for optimization.

> 
> --  eric
> 

Thanks,

Aleksey