On Sunday 17 February 2002 12:42, Ian wrote:

> > I see no harm in this, although I still think that it shouldn't be
> necessary,
It shouldn't be necessary.
> why would an insertion succeed but the data not be propagated
> into the network?  
I think that there are cases where the data can be written to local data 
store before the final outcome of the insert is known.  Maybe this is fixed 
with the new data store code.

Even if the data was written somewhere in the network, there is no guarantee 
that it ended up in a place that makes sense routingwise.  The network used 
to be so small that this didn't matter.  Now it does.

> What realistic circumstance could result in the data
> not being found when the local data store is ignored after a successful
> insertion?  
If it fell out of the network or landed in a ridiculous place routingwise.
>Even if the network was horribly split, the request would
> simply be sent to the same node to which the insertion was sent, and thus
> it should definitely be found.
In a perfect world yes.  We do not live in such a world.   The routing is 
suboptimal. The problem is that most useful nodes are overloaded.  Right now 
my node is QueryRejecting all but ~=1500 of the 10000+ requests it receives 
an hour.

>
> Perhaps someone could do some experimentation to see whether this is
> happening or not.
Insert a 10M splitfile. Wipe your data store and see how many of the pieces 
you can get back.  

--gj

-- 
Freesite
(0.4) freenet:SSK at npfV5XQijFkF6sXZvuO0o~kG4wEPAgM/homepage//

_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://lists.freenetproject.org/mailman/listinfo/devl

Reply via email to