RE: iSCSI: Flow Control

[Robert Snively <rsnively@Brocade.COM>]

>  
>  It does not matter how from where you send the data on the wire.
>  If you have a long wire and you want to cover the latency you will
>  send data as soon as you can and then commands get stuck  behind.
>  
>  And nobody is suggesting you should park the data on the NIC card if
>  you know better.

Don't you simply provide a new command whenever you have a chance
between data transfers?  This could simply be done in any of the
single or multiple connection cases.  Commands should always be
going into a separate space of "unsolicited" actions, while data
should always go into the solicited (or in the case of not using
a ready to transmit signal, into the reserved unsolicited) buffer
space.  Then nothing gets stuck behind anything, except for short-term
serialization timing and perhaps for the duration of serializing a set
of frames for an information unit.

Turn around latency is a second order performance effect for many
configurations, after physical access latency of the peripheral storage
device.  Throughput is maximized because there are multiple threads of
queued commands to multiple logical units.