[Iscsitarget-devel] Clustered target

4 views
Skip to first unread message

Kushnir, Michael (NIH/NLM/LHC) [C]

unread,
Mar 28, 2012, 11:57:46 AM3/28/12
to Iscsitar...@lists.sourceforge.net

Good afternoon,

 

I’m an avid DRBD user. Recently, there was a lot of talk on the DRBD mailing list about DRBD Dual-Primary with iSCSI and multipath. The problems with such a setup are summarized here: http://fghaas.wordpress.com/2011/11/29/dual-primary-drbd-iscsi-and-multipath-dont-do-that/

 

I was wondering If it is possible to add such a feature to IET? Has anyone ever considered such functionality? How much would such an effort cost if someone were willing to sponsor the feature?

 

Thanks,

Michael

 

Ross S. W. Walker

unread,
Mar 28, 2012, 12:43:31 PM3/28/12
to Kushnir, Michael (NIH/NLM/LHC) [C], Iscsitar...@lists.sourceforge.net
In order to have a cluster-aware iSCSI target there would
need to be significant changes in both the configuration side
and operating side.

One change that pops in my head is the coordination of SCSI
reservations between the targets participating. Reservations
could not be granted or revoked until all members approved
the reservation. This means that all IO operations that
depend on frequest RESERVE/RELEASE operations, such as
clustered file systems would hit a triple latency penalty,
or worse for geographically disperse clusters.

One would also need to coordinate configuration of course!

Then there is handling storage failures and you can see
how things start to get way out of control.

The best use of IET and DRBD is in a fail-over scenario.

Have two servers serving up 50% of the storage needs each
off of top-shelf disks, the other 50% of storage off of
slower but still enterprise quality disks, say SAS and
SATA. Have DRBD replicate the SAS on one server to the
SATA on the other server and use something such as
heartbeat to fail-over the targets and IP addresses from
one server to the other in the event of failure.

Each server has multiple interfaces for doing MPIO to
the target, you just don't do MPIO to an active-active
DRBD device across two servers.

For geographically disperse clusters, follow the same
principal as above, but the primary storage is for
in-site hosts and the secondary is fail-over for the
other site. Tricky part here is how to fail-over the
IPs to the other site...

-Ross
______________________________________________________________________
This e-mail, and any attachments thereto, is intended only for use by
the addressee(s) named herein and may contain legally privileged
and/or confidential information. If you are not the intended recipient
of this e-mail, you are hereby notified that any dissemination,
distribution or copying of this e-mail, and any attachments thereto,
is strictly prohibited. If you have received this e-mail in error,
please immediately notify the sender and permanently delete the
original and any copy or printout thereof.
------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Iscsitarget-devel mailing list
Iscsitar...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/iscsitarget-devel
Reply all
Reply to author
Forward
0 new messages