[openib-general] IB: I don't like what I'm seeing.

Thu Apr 1 00:13:09 PST 2004

On Thu, 2004-04-01 at 11:41, Fab Tillier wrote:
> > -----Original Message-----
> > From: Andrey Slepuhin [mailto:andrey.slepuhin at t-platforms.ru]
> > Sent: Wednesday, March 31, 2004 11:26 PM
> > 
> > 2) Ron is right that current IB stacks do not have a good design.
> 
> I beg to differ - I think that *some* of the stacks have a good design.  Do
> note however that good design doesn't equate to good implementation.  A good
> design has the hope of being well implemented.  On the other hand, no amount
> of stellar implementation can help a bad design.  Never the less, in light
> of your comment, I would like to request more specific *design* related
> criticisms.

Ok, may be I didn't use an adequate word, I meant not architectural
design but the design of a code tree. I randomly looked at some places
in the code and they are not looking good to work with. A huge amount of
conditional compilation through all the code, unnecessary code
duplication in some places, etc. Sometimes I was need to think some
minutes to understand how the simple things are implemented. All this is
the source of possible bugs and makes very hard understanding and
modifying the code. I will not point to the exact code fragments, it
will be better if I'll send appropriate patches when SVN tree will be
available :)

>  
> There seems to be a big push to hit the reset button on all these IB stacks
> and start implementing something new.  Better in my mind would be to spend
> some time figuring out the desired design tradeoffs, then picking the stack
> that's the closest (if there is one) and working from there.  Unless we
> spend some time on the design phase, no implementation has much of a hope of
> being adequate.

Oh no, I'm not arguing to redesign all IB stack from scratch. I know
well (based on my own experience in other projects) that this may
improve the things much but will delay the possibility to use the code
for months (though there were positive examples of such rewriting, e.g.
tg3 driver; if everybody was Dave Miller... :))) ). I'm arguing to do a
good code cleanup/audit. Once more please do not blame on me, I'm very
interested that IB stack will be adopted bu kernel community ASAP.

Andrey.

-- 
A right thing should be simple (tm)

-- 
To unsubscribe send an email with subject unsubscribe to openib-general at openib.org.
Please contact moderator at openib.org for questions.