> 
> > > on a slice (or partition) level, where with tapes it was the 
> > > entire device locked out.
> > 
> > That's correct.  I've used the Sequent machines a while back, 
> > and their
> > lock manager had a proprietary protocol which they ran over the SCSI
> > bus.  I'd love a similar type of thing for NetBSD...
> > 
> 
> I wonder if there has been any work done to do IP over SCSI.

You mean other than the se device (scsi/ethernet)?

Not directly, but something like this is on my plans for this year at least
for Fibre Channel.

Have you looked into the GFS (Global File System's) DLM work as well as their
DLOCK (device lock) stuff?

>  I have
> a heartbeat system in my first release of NetBSD clustering, though no
> lock manager.  It all is done with UDP datagrams over redundant networks
> 
> Sequent had chosen to use both SCSI and the NETWORK to send the
> messages, which IMHO is a little paranoid.  And I think it buys a 
> few more problems.  I think the rational was because of the networking
> being streams based and there were a few cases where it was easy to
> dry out the streams on temporary basis which would cause a false
> cluster transition.  Since we don't have streams to content with,
> I didn't really think too much about it when I was doing the heart
> beat stuff.
> 
> It would be nice to have a universal transport protocol sort of 
> like VI was supposed to be (and I guess won't). It's funny to see all
> the talk about SAN's (Storage area networks) moving in the direction
> of using an IP protocol because of distancing and routing.  

That's the eSan story, yes. I'm not convinced it will do win. There are
reasons to do IP over SCSI, but more for bridging/legacy issues than
performance. It'll probably be a race between true interoperability for the
GigEther switches and 64k or larger tcp/ip sizes (with larger MTU sizes)
versus the FC switch vendors getting their E-ports (switch-to-switch
connections) to play with each other and getting all the !@)%$*)$)!$(!$*(
other interop problems for FC h/w to work and getting the OS's to provision
WWN addressing correctly.

-matt