Re: iSCSI: multiple initiator conflicting with target (was Re: iSCSI: response to second login (with same ISID)

[julian_satran@il.ibm.com]

As far as I know this is a "primal" issue with all cooperating systems.
SCSI has attempted to handle it with very granular reserve/release. However
those where never very popular
with vendors and users (except for exoteric systems block level
reserve/release where not implemented) as they carry a high price tag and
performance is quite low.  "Complicated software vendors" have fared
slightly better
as initiators that "cannot cooperate" can in fact and more effectively than
targets - as they have a good handle on the lock semantics.

However this is hardly an iSCSI issue and I suggest we take this off the
IPS List.

Julo

"Sanjeev Bhagat \(TRIPACE/Zoetermeer\)" <sbhagat@tripace.com> on 29-05-2001
03:57:39

Please respond to "Sanjeev Bhagat \(TRIPACE/Zoetermeer\)"
      <sbhagat@tripace.com>

To:   "Patrick Stirling" <pms@veritas.com>
cc:   "'Lakshmi Ramasubramanian'" <nramas@windows.microsoft.com>, "Santosh
      Rao" <santoshr@cup.hp.com>, ips@ece.cmu.edu
Subject:  Re: iSCSI: multiple intiaitor conflicting with target (was Re:
      iSCSI: response to second login (with same ISID)




Hello Patrick,

Since the initiators are two different entities working completely
independent of each other they cannot co-ordinate with each other. iSCSI
Target also sees these two different initiators in different sessions as
seperate entities. In the current implementation of iSCSI I guess it is not
possible for two different initiators to know about each other and so it is
difficult to resolve this conflicting situation.

The answer to this problem might be lying in SAM-2 specifications. May be I
am missing something there. Can somebody comment?

Sanjeev

>
> On Mon, 28 May 2001, Sanjeev Bhagat (TRIPACE/Zoetermeer) wrote:
>
> > Well the conflict of writing to same sector can still be avoided by
> > implementaion of some locks in the target but I wonder what will happen
in
> > this case where 2 inititators are connected to a target.
>
> > From: Lakshmi Ramasubramanian [mailto:nramas@windows.microsoft.com]
> >
> > Even if the target is capable of supporting multiple initiators, won't
> > it cause problems  with devices such as disk - say, filesystem from the
> > two initiators' side attempt to write
> > to the same sectors (even not intentionally) would cause data
> > corruption. Is iSCSI layer suppose to guard against this type of
> > device sharing?
>
> Umm, am I missing something here?
>
> If 2 initiators want to access the same SCSI device, they must coordinate
> to avoid clashing with each other. It's not the device's responsibility
to
> prevent this. This is one reason several companies have complicated
> clustering software.
>
> patrick
> Patrick Stirling
> VERITAS Software (vendor of complicated clustering software!)
>