Hi Anil, To put the story short in my interpretation - others could correct me if I am wrong:
Implementations - including software switches - differ in the precision they can provide, up to the ~100ms range. So basically a granularity of seconds is what can always be guaranteed. (Even though the reported statistics are in ns, any precision beyond the ms range is "best effort" - depending on the actual implementation you use.) Also, the primary use for flow timeouts is to clean up flow entries, and not to (ab)use them as a high precision timer on the switch, so a granularity of seconds is sufficient. The technical aspect of the problem is that if we went below this granularity, the protocol would become more complex. I.e., the controller would have to tell not only the timeout value, but also the desired granularity for it. For example, the protocol would have to make a distinction between a timeout of 4s, and a timeout of 4.000s. Where the latter should be rejected (with bad timeout) by switches not providing that granularity. Nevertheless, if you have a valid use-case for sub-second flow timers, please share it with us, so we can add it to those discussions. Regards, Zoltan. ________________________________ From: openflow-discuss-boun...@lists.stanford.edu [mailto:openflow-discuss-boun...@lists.stanford.edu] On Behalf Of Anil Vishnoi Sent: Thursday, June 07, 2012 10:16 AM To: Dan Talayco Cc: openflow-discuss Subject: Re: [openflow-discuss] Idle_timeout & Hard_timeout time granularity to seconds Hi Dan, I am not able to open any of these URLs. It throws permission violation (It seems that you have tried to perform an operation which you are not permitted to perform.). Can you give me any other link where i can see these stories/bugs. Thanks Anil On Thursday 07 June 2012 01:02 PM, Dan Talayco wrote: On Thu, Jun 7, 2012 at 12:19 AM, Dan Talayco <dan.tala...@bigswitch.com<mailto:dan.tala...@bigswitch.com>> wrote: First, note that the precision with which durations are reported is in nano-seconds in flow stats and expiration messages. This provides support for more accurate measurements using this info. There has not been much of a call for increasing the granularity of time-out settings below 1 second. Are you considering using timeouts of less than a second on flows? Zoltan politely pointed me to these: https://www.opennetworking.org/bugs/browse/EXT-86 https://www.opennetworking.org/bugs/browse/EXT-57 -Dan On Thu, Jun 7, 2012 at 12:10 AM, Anil Vishnoi <vishnoia...@gmail.com<mailto:vishnoia...@gmail.com>> wrote: Hi Team, Is there any specific reason / restriction to keep the idle_timeout & hard_timeout values to second level granularity and not milli/micro/nano seconds? Anil _______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu<mailto:openflow-discuss@lists.stanford.edu> https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
_______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
