Re: iSCSI: ImmediateData

[sandeepj@research.bell-labs.com (Sandeep Joshi)]

One could tweak the advertised queue size (max-expCmdSN)
if the target does not have enuf memory.  It would avoid 
the R2T delay on every command.

-Sandeep

> [ Attached is an exchange with Julian mid-last week. ]
> 
> Julian,
> 
> I once again recommend that the default for ImmediateData
> be "no".  I would argue that the most simple case should be
> enabled by default - just what we did for InitialR2T -
> not the one that demands more memory for unsolicited
> immediate data.
> 
> Comments?
> --
> Mallikarjun
> 
> Mallikarjun Chadalapaka
> Networked Storage Architecture
> Network Storage Solutions Organization
> MS 5668 Hewlett-Packard, Roseville.
> cbm@rose.hp.com
> 
> - I realized (during plugfest) that ImmdiateData's default is
> "yes".  I recommend making it "no", keeping simple-minded targets
> in view.  This issue is more serious if those simple targets
> do not do a lot of text negotiation, when they will be stuck
> with having to support a DataPDULength worth of immediate data.
> This is also in contrary to the conservative spirit of the default
> for InitialR2T (whose default is "yes").
> 
> Please feel free to take it to ips if you think it's a change that
> warrants comments from a  wider audience.
> 
> +++ I did not hear any comment from the plugfest on this.
> The rational behind it is that simple software implementions benefit the
> most out of it
> and the default buffer is small ++
>