Inline. On Nov 14, 2010, at 2:55 PM, kk yap wrote:
> Hi, > > Comment inline. > > Regards > KK > > On 14 November 2010 09:51, Rohit Manohar <rdman...@ncsu.edu> wrote: >> Hi >> >> Please ignore the previous mail. >> >> We have decided to implement load distribution based on counters maintained >> by a switch for a flow. When we install a flow in a switch, we will set some >> 'hard_timeout_value'. After the specified time, we expect the switch to >> return a 'flow_expired_event'. I have also found 'flow_removed_event' which >> is very similar to the expired one. > > The flow_removed_event is the more update-to-date version of the > flow_expired_event. The main difference is that flow removed is > triggered even when the flow is removed explicitly instead of with a > timeout. > >> What is the difference between the two? >> Do these events return 'packet_count' in the handler. I am not able to find >> the code for these events. Could you provide with a path? IIRC, the handler for flow_removed events just gets the event object itself. These are C++ events which are converted to Python by pyrt, so you can find what they contain by looking in pyrt.cc (or just print dir(event) in the handler). The event object does have the packet count on it. > Sorry, I parse my own packets in C/C++ and I am currently away for my > development machine for a couple of weeks. Can someone point Rohit to > the correct API? > >> >> Regards, >> >> On Sun, Nov 14, 2010 at 11:16 AM, Rohit Manohar <rdman...@ncsu.edu> wrote: >>> >>> Hi KK >>> >>> Thanks for the suggestion. In fact we were planning to build our system on >>> similar lines. The idea is to divide the execution time into 'epochs'. At >>> the start of each epochs, we plan to read from the number of bytes from each >>> flow and aggregate them to determine the total flow on an output port. Based >>> on these, numbers we will redistribute the flows. Since, events drive >>> everything in NOX, we will need an event 'start_epoch', based on a timer to >>> trigger this calculation. I read your post on nox-dev regarding adding an >>> event into NOX. Is such a timer-based event already in NOX, or we will have >>> to define one? If we have to, could you through some light on the procedure >>> and pitfall of doing one? >>> >>> Regards, >>> >>> On Sun, Nov 14, 2010 at 2:17 AM, kk yap <yap...@stanford.edu> wrote: >>>> >>>> HI Rohit, >>>> >>>> I do not like you can implement this using OpenFlow in a pure-datapath >>>> manner, because OpenFlow does not match on the counters. However, I >>>> believe you can approximate this by polling the counters and changing >>>> the flow rule when the counters satisfy some conditions. Of course, >>>> this results in some delay in the change. >>>> >>>> Regards >>>> KK >>>> >>>> On 13 November 2010 17:26, Rohit Manohar <rdman...@ncsu.edu> wrote: >>>>> Is is possible to define multiple actions for a single flow based on >>>>> counters for that flow. The idea is to choose different output ports >>>>> based >>>>> on number of packets encountered in that flow. >>>>> >>>>> Regards, >>>>> -- >>>>> Rohit Manohar >>>>> Graduate Student >>>>> North Carolina State University >>>>> Raleigh, US. >>>>> >>>>> >>>>> _______________________________________________ >>>>> nox-dev mailing list >>>>> nox-dev@noxrepo.org >>>>> http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org >>>>> >>>>> >>> >>> >>> >>> -- >>> Rohit Manohar >>> Graduate Student >>> North Carolina State University >>> Raleigh, US. >>> >> >> >> >> -- >> Rohit Manohar >> Graduate Student >> North Carolina State University >> Raleigh, US. >> >> > > _______________________________________________ > nox-dev mailing list > nox-dev@noxrepo.org > http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org _______________________________________________ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
