[ale] Fileserver/filesystem replication

[Stephan Uphoff]

Take a look at:

http://www.usenix.org/events/hotos03/tech/full_papers/henson/henson_html/hash.html


to see why this is a bad idea.


	Stephan


> All,
> 
> On Wed, 2003-12-31 at 04:13, Bob Toxen wrote:
> > Ya want real-time replication.  I'll give ya real-time replication.
> > 
> > 1. Build your ext3 or ReiserFS on top of Linux software mirroring.
> > 2. Make the second (mirror) disk a network disk device.
> > 3. Ensure that writes are not acknowledged until they percolate through
> >    to the remote real hard disk.
> 
> 	When I was working for Connex (a startup NAS box company absorbed by
> Quantum, alas), we had a guy in Germany doing just about what you
> describe above, with another wrinkle.  He was using the Network Block
> Device (see the kernel docs for NBD) as one 'disk' of the mirror, but he
> took it one step further.  As a block came in to be written, he
> performed an MD4 (maybe MD5) checksum of it, and sent that across first,
> along with the block number.  If the data was already there (I don't
> have or recall any stats on it), then he was done.  If not, the NAK
> would cause the whole block to be sent.  Higher latency, but as I
> remember, it worked pretty well.
> 
> 	Just FYI.  I don't know if it was ever released, or even if it ever saw
> any use outside our "lab".
> 
> -- 
> kernel, n.: A part of an operating system that preserves the
> medieval traditions of sorcery and black art.
> 
> Danny
> 
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://www.ale.org/mailman/listinfo/ale
>