HEADS UP - devfs integration in master this weekend

[Matthew Dillon]

Alex's DEVFS SOC work will be integrated into the git master this
weekend.  The development system may be unstable for a week or so after
the integration.  Anyone upgrading after the work is committed must
take care to upgrade both their kernel and the world.

We are also on-track for getting a new release of gcc into the tree
(from Simon) and I hope the iSCSI client too (from Antonio).

Mount-by-serial-number support and /etc/devtab support (mount-by-label,
based on serial numbers) will be added to devfs after it is integrated.
This is big deal too as it gets rid of the problem of device renumbering
if one has a lot of hard drives... something that is easy to do these
days with E-Sata and Port-Multiplier enclosures.  Not to mention iSCSI.

The 2.4 release is still on track and scheduled for September.

-Matt
Matthew Dillon 
dil...@backplane.com

Re: HEADS UP - devfs integration in master this weekend

[elekktretterr]

I cant wait. I think devfs is really important, with so much static nodes
in /dev its really a mess.

Just a side question, when HAMMER becomes network aware (disks
spanning/replicating over network in a cluster), will it be possible to
mount disks on other DragonFly servers like mount_hammer
/dev/disk_on_the_network /mnt/mydisk and it will just know that
disk_on_the_network is a disk on a different server that is part of the
cluster?

This will be the greatest release yet no doubt.

Cheers,
Petr

Re: HEADS UP - devfs integration in master this weekend

[Matthew Dillon]

:I cant wait. I think devfs is really important, with so much static nodes
:in /dev its really a mess.
:
:Just a side question, when HAMMER becomes network aware (disks
:spanning/replicating over network in a cluster), will it be possible to
:mount disks on other DragonFly servers like mount_hammer
:/dev/disk_on_the_network /mnt/mydisk and it will just know that
:disk_on_the_network is a disk on a different server that is part of the
:cluster?
:
:This will be the greatest release yet no doubt.
:
:Cheers,
:Petr

Yes, certainly anything made available via iSCSI will be accessible
that way even without the cluster work.  DEVFS, /etc/devtab, and
being able to access drives by serial number is the crucial enabler.
Antonio may disagree but I think he's close enough with his iSCSI
port that we can finish it off next month.

The cluster work itself will operate at and above the filesystem
layer and is far more ambitious.  It is still at least a year away.
Of the major components the HAMMER piece is done, and the replication
piece has working infrastructure (HAMMER's mirroring feature).  Left
to do is the quorum commit protocol, RPC mechanism for communication
between nodes, and what will eventually be a very complex cache
coherency piece.

Without getting into too much detail, being able to extend the
master-slave replication that HAMMER has now to master-master
replication requires the cache coherency piece as the mechanic which
prevents conflics in the replication streams.  With that in place
HAMMER's mirroring feature can be used to fan-out the replication
of changes with no topological limitations.

-Matt
Matthew Dillon 
dil...@backplane.com