Re: iSCSI - change proposal LUN field definition on every PDU

["Mallikarjun C." <cbm@rose.hp.com>]

> IMO, the LUN knowledge is best kept out of the SCSI transport as far as
> possible.

I agree.  

I also agree with Paul Koning about the perf path implications for
software implementations.

In addition, more significantly, I am concerned about the consistency 
checking aspects of this change.  Apart from the performance costs 
incurred because of checking, what if the LUN doesn't match the 
active ITT?  In general, more errors (and possibly Reject reasons).

Lastly, it appears to me that the general philosophy of relying only
on ITT for instantiated tasks is not applied here (given that ITT
is unique per session across all LUNs).

To summarize, IMHO, these disadvantages outweigh the benefit of 
stateless logging of LUN traffic.
-- 
Mallikarjun 


Mallikarjun Chadalapaka
Networked Storage Architecture
Network Storage Solutions Organization
MS 5668	Hewlett-Packard, Roseville.
cbm@rose.hp.com


Santosh Rao wrote:
> 
> Julian,
> 
> Can you explain further the extent of PDUs that are intended to be
> affected by this change ? Also, can you clarify why it is not a problem
> for other SCSI transport protocols like FC where the LUN field is not
> included in every FC IU ?
> 
> IMO, the LUN knowledge is best kept out of the SCSI transport as far as
> possible. All LUN level logging/tracing belongs to the SCSI ULP, which
> is given a LUN context on each call made to it from the LLP.
> 
> At the SCSI LLP layer, tracing/logging as well as state information
> should be at the I-T nexus level (session).
> 
> I don't (yet) see a sufficient justification for this change. Perhaps,
> more information on the motivation for this request may change that.
> 
> - Santosh
> 
> Julian Satran wrote:
> >
> > Dear colleagues,
> >
> > A colleague interested in instrumentation approached me with a question
> > about stateless logging of specific LU activity.
> > With the current iSCSI PDU formats this is not possible.
> > We have consistently avoided having fields that are redundant and will
> > need consistency checking.
> > However I think we should consider including the LU field in all PDUs that
> > are referencing a specific LU to simplify this type of instrumentation (as
> > we did with the direction bit in the op-code).
> >
> > As I am already in "count-down" mode for version 09 I would appreciate
> > your comments ASAP.
> >
> > Julo
> 
> --
> ##################################
> Santosh Rao
> Software Design Engineer,
> HP-UX iSCSI Driver Team,
> Hewlett Packard, Cupertino.
> email : santoshr@cup.hp.com
> Phone : 408-447-3751
> ##################################