Home of the L4 community
         Home )   About L4hq )  
The following is Gernot Heiser's announcement of a bet that emerged from the discussion about the exception handling model in late 2001.


The deadline for the exception handling discussion expired on Wed. As
no agreement has been reached, the ball is now in my court. Rather than
behaving like a biased random-number generator I'll try to give a
rendition of King Salomon ;-)

My judgement is as follows: We'll stick with the IPC model. Not
because I'm convinced by the elegance/cleanness arguments, but because
I think it poses a research challenge. We're likely to learn more by
trying it than by not trying. In particular the argument brought
forward by Kev that the model may have limitations which prevent its
universal usage will need to be examined.

To provide an incentive to explore the issue, and make to things
interesting, I'm imposing a few extra conditions.

The decision will be taken as a "trial" one, and will be accompanied
by an open challenge to prove it wrong. The proponents of the IPC
model will deposit EU25.- for a prize which can be claimed by whoever
can come up with a reasonable "real life" benchmark that demonstrates
a >=5% slowdown resulting from the IPC model of exceptions. The rules
of the challenge are:

 - It is open to any user of L4. The challenge and its rules will be
   published together with other comments on the proposed new API.
 - It runs until one year after release of the first conformant kernel.
 - Everyone claiming to have a realistic example which performs at
   least 5% worse with the IPC model than with a well-implemented
   same-context mode will contribute EU5.- to the prize and has a
   chance to win the lot.
 - The 5% may be reached on _any_ modern architecture (or IA-32).
 - The proponents of the IPC model are free to tune the IPC
   implementation to reduce the overhead (as long as it still conforms
   with the original spec).
 - If by the end of the year at least one such a claim is made and
   accepted as valid, the IPC model will be abolished. (It may be
   abolished sooner.)
 - If there is more than one such accepted claim, the one with the
   highest overhead wins, if there is only one, that one wins,
   otherwise the original proponents of the IPC model win.
 - The winner will receive the prize in the form of wine, champagne,
   or whatever their favourite poison may be (as long as it's legal in
   Germany and the residence of the winner).
 - Gernot will be the judge.


PS: Being the neutral judge I don't consider myself one of the
"proponents", but I'll contribute my EU5.- anyway.

The first conformant kernel (L4Ka::Pistachio) has been released on May 2 2003. The jackpot currently contains € 15.