Re: AoE - ATA over Ethernet - comments?

Don't confuse me with an expert on this subject, but I understand that 
Coraid's intent for AoE is to allow a stand-alone HDD [i.e. EtherDrive 
blade] to serve as a network storage appliance. Put multiple such HDDs in 
a rack, add AoE driver software and distributed filesystem support [e.g. 
GFS] on the client, and call the rack a storage array. Each EtherDrive 
blade is effectively an HDD with an Ethernet interface and server software 
that shuttles ATA commands and responses between the IDE registers and the 
network.

Coraid has GPL'd a virtual EtherDrive blade called "vblade" which runs on 
Linux and FreeBSD platforms. A vblade exports storage [e.g. a disk, a 
partition, or a file] such that from the perspective of an AoE client it 
looks exactly like an EtherDrive blade - including the fact that it must 
be configured with shelf and slot numbers. Under the covers, instead of 
writing directly to the IDE registers, a vblade uses standard ioctl(), 
lseek(), read(), and write() system calls. If you wanted to use vblade to 
emulate the presence of multiple EtherBlades, then multiple instances of 
vblade must be run. Of course, you could re-write vblade to do whatever 
you'd like it to do.

... Dave





Rudolph Bott <[email protected]>
Sent by: [email protected]
12/09/2005 05:49 AM
Please respond to rb; Please respond to Enhanced Network Block Device

 
        To:     Enhanced Network Block Device <[email protected]>
        cc: 
        Subject:        Re: [ENBD] AoE - ATA over Ethernet - comments?


[email protected] wrote:

> - AoE is a very simple protocol - perhaps much simpler than NBD - 
perhaps 
> too simple for your needs?
Since it is not tcp/ip- (and port-) based, can you "publish" multiple 
devices? (e.g. compared to (e)nbd, where you have to run a dedicated 
server process for each device)
And what about the integration part, does it rely on userspace tools? We 
are using an initrd image over here which cares about starting the 
nbd-client, mounting /dev/nb0 as root device etc.
This kind of setup requires some changes to the standard (debian-) 
system, because you need a patched version of killall5. Otherwise our 
nbd-client process will get killed during shutdown/reboot by 
/etc/init.d/sendsignals, leading to a system crash because the root 
device disappears before init has finished calling the shutdown scripts
To make things easier and to avoid those nasty work-arounds it would be 
nice to look at something in-kernel-like (but nfs is definatly not an 
option).


-- 
Mit freundlichen Gruessen / with kind regards
  Rudolph Bott

-------------------------------------------------------------------
Megabit Informationstechnik GmbH  Karstr.25  41068 Moenchengladbach
Tel: 02161/308980   mailto:[email protected]       ftp://megabit.net
Fax: 02161/3089818  mailto:[email protected]   http://megabit.net
-------------------------------------------------------------------
_______________________________________________
ENBD mailing list
[email protected]
http://lists.community.tummy.com/mailman/listinfo/enbd