On Sat, 2004-12-18 at 12:55, Roland Dreier wrote:
> Surely link width and/or speed can't change without the port state
> changing, can they? As I understand it, the link layer can't
> renegotiate this sort of thing without bringing the link down. In
> which case ULPs only need to refresh their rate information when they
> get one of the existing port state change events.
Yes, port state change is sufficient to take care of this.
> (If the link rate can change without any notice, then static rate for
> RC is meaningless since a link could change from 4X to 1X without
> disturbing existing connections).
Keep in mind that unless one is monitoring all of the intermediate port states, then this can change without it being noticed by the CA port if does not change. Same would apply if a switch fabric were reconfigured such that a LID took a new path through the fabric and the rate was impacted by either link width change or signaling rate change. Static rate needs to understand the entire end-to-end path and whenever it is updated to be effective. For IP over IB using UD, there are many paths which can be taken so it is not clear what would be an appropriate IPD if one path the data rate is 500 MB/s and on another it is 6GB/s. Should this be dominated by the lowest path speed or should one pursue VL arbitration to segregate this type of workload based on what the endnode believes is the proper QoS? IPD does not really work for UD in the end and our intention with the technology was to avoid trying to bandwidth manage UD other than through VL / switch port arbitration policies.
Mike
_______________________________________________ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
