[cap-talk] - Bellizzomi - Capabilities and Shapiro's focus, Coyotos, etc.

Valerio Bellizzomi devbox at selnet.org
Wed Nov 29 14:24:16 CST 2006


On 29/11/2006, at 8.41, Mark S. Miller wrote:

>Karp, Alan H wrote:
>> While you can't prevent wall banging, you can prevent wall listening by
>> removing all forms of indeterminacy, such as access to the system
clock.
>> Any process that is deterministically replayable meets this criterion.
>> Did I get that right, MarkM?
>
>Yes thanks!
>
>In addition, once such processes are deafened, this can contribute
towards 
>alleviating wall banging in an indirect and imperfect way:
>
>The main wall banging channel is variable resource use. The main defense 
>against such wall banging is preallocation of worst-case resources, and
>then 
>wasting the slack -- the difference between actual and worst-case
resource 
>needs. Since typical resource needs can often be orders of magnitude less
>than 
>worst case, this waste can be quite costly.
>
>However, if one has a bunch of deafened processes that could do useful
>batch 
>work, one can run these in the slack. As deafened processes, they can't
>sense 
>how much slack they're getting, or how this slack is spread over real
time.
>
>By "batch" above, the key feature is that no one needs an answer from
>these 
>promptly, so these answers can be delayed and aggregated. Revealing which

>answers are available at some future time does reveal some covert info,
>but 
>hey, I said it was imperfect.

Well I said that "One thing that could be avoided inside the software to a
certain extent, is the possibility of wall-banging".
Ok, let's say that wall banging can be mitigated to some extent.

val

>
>-- 
>Text by me above is hereby placed in the public domain
>
>     Cheers,
>     --MarkM





More information about the cap-talk mailing list