On Dec 5, 2013, at 11:20 AM, Zoltán Lajos Kis <zoltan.lajos....@ericsson.com> wrote: > What might cause confusion (if there is any) is that the original OF 1.0 > implementation provided extension messages - OFP_EXT_QUEUE_MODIFY and > OFP_EXT_QUEUE_DELETE - for managing queues. See e.g.,: > http://yuba.stanford.edu/git/gitweb.cgi?p=openflow.git;a=blob;f=utilities/dpctl.c;h=0406d9b5f21557a9450bd573755d17319723a3ea;hb=HEAD#l247
the Indigo-1 switch firmware also supports those extensions for in-band hardware queue control [1]. in addition, the firmware adds support for max bandwidth (rate-limit) as well as guaranteed minimum bandwidth: https://github.com/bigswitch/iods/tree/master/indigo-core/openflow pronto01# dpctl --help ... Queue Ops: Q: queue-id; P: port-id; min/max: bandwidth in 0.1% (1-1000) add-queue SWITCH P Q [min [max]] add queue [min bw [max bw]] mod-queue SWITCH P Q [min [max]] modify queue min/max bandwidth To disable bandwidth settings, use 0 for min and/or max del-queue SWITCH P Q delete queue dump-queue SWITCH [P [Q]] show queue info ... note that there is one important trick to using these queues: the first queue you create will be used as the "default" queue (that is, rules with an action "output:N" will go to the first queue you create on port N). to simplify things, I find it helpful to run a basic script like the following on switch startup: https://gist.github.com/adferguson/7863782 I can report that we have used these extensions successfully for hardware QoS in our PANE controller: https://github.com/brownsys/pane/blob/nib-rewrite/src/ControllerService.hs#L125 for those who wish to test controllers which use the rate-limit extension as well, I have a fork of the original OF 1.0 implementation with the same: https://github.com/brownsys/openflow hope this helps, Andrew [1] http://www.openflowhub.org/display/Indigo/Indigo+-+Open+Source+OpenFlow+Switches+-+First+Generation _______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss