> Message: 3
> From: "Maarten Geurts op H0tmeel" <[EMAIL PROTECTED]>
> To: [EMAIL PROTECTED]
> Date: Tue, 23 Apr 2002 10:48:28 +0000
> Subject: [freenet-tech] FASD Fuzzy Search.. deletes?
> Reply-To: [EMAIL PROTECTED]
> 
> The document (well ian's pdf copy....) describes that I "describe
FASD, a
> fault-tolerant adaptive, scalable, distributed search engine inspired
by
> Freenet and designed with Freenet in mind. "
> 
> I think  the deletes (culling) of metadata a problem. First "whenever
a
> node
> suspects that 'data' no longer exists". should be more specific (after
a
> "404 error"?)
[Amr:] Yes, a node suspects data no longer exists when it requests the
document associated with the metadata key and receives a request
unsuccessful message.
 wouldn't this in effect create a lot of culling floods for
> impopular data? Suppose by a LRU method both metadata and data would
be
> out
> of the network. Now however some node finds the metadata but fails to
find
> the data. 

[Amr:] This type of synchronization issue is precisely the reason why we
need to delete metadata keys.  The reason each node should verify that
the metadata-cull is legitimate is to avoid censorship of keys or the
actual reality that the document is just out of reach of the initial
requestor.
what would network wise be more efficient: just drop the
> metadata
> or engage a protocol to delete a key across a lot of nodes? My feeling
is
> that this would create a lot of trafic that would only result in
> propangation of data that is impopular. Since popular metadata for
> impopular
> data would have a lot of effect on the distibution of that data this
will
> make it impossible to destroy this data.
[Amr:] I don't understand how the metadata key-cull method will
replicate unpopular data?  If there is a metadata key-cull and it does
result in propagating the document that means a user tried to request
the document and failed (thus the document should be wore widely
replicated) or the user tried to censor the metadata key (thus the
document should be replicated)

Replication of the document only occurs if it found. In this case it
should be replicated because a user requested it and the request failed.
> 
> This is positive for the survivabilty of the data but clashes with the
> limited resources of the entire network.
> 
> This sound a lot like "instead of deleting data in nodes store send it
to
> the neighbour".
[Amr:] Not true.  The policy is simply "Before deleting metadata key
make sure the document is really gone"
Finally, if the key-cull method really does generate too much traffic
then you just have nodes choose to verify the cull request with some
probability.  The higher the probability the less likely is an erroneous
cull.

> 
> Just my 2 cents.
> 
> Maarten.
> 
> 
> 
> 
> 
> 
> _________________________________________________________________
> Send and receive Hotmail on your mobile device: http://mobile.msn.com
> 
> 
> 
> 
> --__--__--
> 
> _______________________________________________
> freenet-tech mailing list
> [EMAIL PROTECTED]
> http://lists.freenetproject.org/mailman/listinfo/tech
> 
> 
> End of freenet-tech Digest


_______________________________________________
freenet-tech mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/tech

Reply via email to