[Gluster-devel] DHT/Replication enhancements

[Frank Clements]

Hello,

I see that the DHT translator doesn't cope well with node removal and
additions (updating conf files on many servers) might be problematic.  Also
the only real method of "reliability" is with replication.

So I got to thinking of how P2P networks use DHT to locate files and deal
with node failure/removal/additions and wondering if any of the developers
have ever considered implementing something like this - distributed tracker
in p2p parlance.   Another issue with the replication part is the fact that
you have to fully duplicate the data on, ideally, more than 2 nodes which
seems like a waste of space.  Using something like erasure coding (think
raid 5/6) could provide the reliability piece of this.  I recently read an
interesting paper on replication vs. erasure coding which seems to make
sense.   

http://oceanstore.cs.berkeley.edu/publications/papers/pdf/erasure_iptps.pdf

Has anyone considered anything like this?  Would it even be possible, or am
I completely missing the boat!?   


Frank Clements

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 3568 bytes
Desc: not available
URL: <http://supercolony.gluster.org/pipermail/gluster-devel/attachments/20090723/1163fc56/attachment-0003.bin>