Re: [Iscsitarget-devel] IET, WinTarget, and PyX Comparison

[Nicholas A. B. <na...@li...>]

On Mon, 2006-02-27 at 12:40 -0800, Rik Herrin wrote: 
>
> --- "Nicholas A. Bellinger" <ni...@sb...> wrote:
> 
> > Greetings All!
> > 
> > Registered iSCSI logical units provided to the host
> > OSes SCSI subsystem
> > speaking ERL=2 can preform session continuation
> > within the context of
> > its own iSCSI Nexus.  
> 
> So it's mainly useful for session continuation (which
> is a nice feature).  But does it affect data integrity
> in any way?  Would using plain erl=0 with HeaderDigest
> and BodyDigest give you everything that erl=2 gives
> you without session continuation, or could erl=2
> prevent certain data corruption that erl=0 combined
> with HeaderDigest and BodyDigest can't protect you
> from?
> 

Data integrity by way of 32-bit application layer checksums are not
affected by ERL.  When iSCSI is used across other network interconnects
such as iSCSI/SCTP and iSER/iWARP & iSER/IB  these application layer
digests are disabled and data integrity checks are left up to the
transport layer and/or hardware.

So while ERL=2 does not add any additional data integrity protection, it
does add the ability to recover from connection exceptions without
affecting other communication paths, and hence service requests sooner
than if all connections had to be stopped/restarted.  This starts to
become important as the iSCSI connection count and communication path
failure increases.  In these cases, ERL=2 is used as highly efficent
internexus multipath mechinism that is done independently of the
underlying network protocol(s).

> > 
> > Oh btw, core-iscsi is alive
> 
> It would be nice if it got more publicity.  I was
> under the impression (as I believe a lot of other
> people are as well), that it was discontinued.
> 

One of the areas of work recently done with Core-iSCSI has been getting
the more advanced features (namely MC/S) validated with 3rd party iSCSI
targets.  With mailing list members help, I believe Core-iSCSI has
become the world's first Linux Initiator to prove MC/S interopt across
multiple iSCSI Target implementations.

The Core-iSCSI project is going to continue to focus on these types of
efforts, as well as continue to make the software useful to as wide of
an audience as possible, and add additional bits that make iSCSI SANs
(iSNS for example) easier to administrate and use on a daily basis.
Also, I believe linux-iscsi.org will help getting potential users
interested in Linux/iSCSI.

> > , it is currently in
> > stable release and
> > accepting feature requests.  Please check out the
> > following wiki entry
> > for more information:
> > 
> >
> http://www.linux-iscsi.org/index.php/Core-iscsi_Interopt
> > 
> > Nicholas A. Bellinger <ni...@sb...>
> 
> Nicholas, have you performed any benchmarks between
> IET and PyX or between PyX and other iSCSI targets? 
> If so, could you share them with the rest of us,
> including the configurations used?  Finally, I
> couldn't find much about PyX on your site:
> (http://www.sbei.com/index.php/products/iscsi/)  Does
> it come with a management interface?  How is it priced
> (per connection like WinTarget or per server)?
> 

There where 10 Gb/sec Internexus+Outnexus numbers published
with Neterion NICs and the PyX/SBEi Target in combination
in Linux v2.6 with dm-multipath and Core-iSCSI last year:

https://www.redhat.com/archives/dm-devel/2005-April/msg00013.html

There was also a 10 Gb/sec ERL=2 whitepaper published by
Force10 Networks, Neterion and PyX from last year as well:

http://www.force10networks.com/products/iSCSI_10GE.asp

I believe there have been MC/S 4x 1 Gb/sec numbers published
on x86_64 as well.  You might want to check google for these.

Thanks for your interest!

-- 
Nicholas A. Bellinger <na...@li...>