Btw I meant, it would block or drop messages if you use a HWM that is low (but not 0). 0 should give you unlimited limit, but you will eventually block or drop messages (tcp buffers filling up, etc). Better load balancing will be achieved by using a router socket
On Fri, Jan 18, 2013 at 10:53 AM, Harsh Doshi <[email protected]> wrote: > 0 would mean no limit. And it would either block or drop messages (I > think). > You'll need a router socket > > > On Fri, Jan 18, 2013 at 9:25 AM, Nishant Mittal <[email protected]> wrote: > >> i see.. that might explain why DLR is blocking.. so do you think if i set >> the SND buffer to 0 on the dealer.. this will work? effectively giving me a >> load balanced broker? >> >> >> On Fri, Jan 18, 2013 at 11:10 AM, Stuart Brandt <[email protected]>wrote: >> >>> Your DLR has a send_hwm too, so it's going fair queue against its >>> outbound queues (which might not be full despite receiver not accepting >>> more). The mentioned load-balancing-message-broker or one of the reliable >>> request-response patterns in section 4 will probably serve you better than >>> trying to control routing by adjusting HWM. >>> >>> >>> On 1/18/13 10:58 AM, Nishant Mittal wrote: >>> >>> aah, i remember that now.. i'll look into it again. >>> my hope was DLR would "try" to fair queue but if the worker was not >>> accepting any more msgs.. DLR would try the other worker.. but its not >>> doing that. >>> >>> thanks >>> >>> On Fri, Jan 18, 2013 at 10:49 AM, Andy Ballingall TF < >>> [email protected]> wrote: >>> >>>> >>>> On 18 January 2013 15:14, Nishant Mittal <[email protected]> wrote: >>>> >>>>> I have 3 workers (REP) connected to a DLR socket. each worker takes a >>>>> msg, processes and responds... however, as the DLR socket fair-queues.. >>>>> requests are sent like this... >>>>> 1st -> 1st worker >>>>> 2nd -> 2nd worker >>>>> 3rd -> 3rd worker >>>>> 4th -> 1st worker >>>>> >>>>> problem is if the 1st worker is still busy with the 1st request the >>>>> 4th request waits... even if the 2nd and 3rd workers are free. I am >>>>> "guessing" its because of the receive buffer on the REP socket.. >>>>> >>>>> is it possible to fix this by setting the ZMQ_RCVHWM to 0? >>>>> >>>>> >>>> I'm not sure if it will help your use case, but have you read through >>>> the load balancing example in the guide? >>>> >>>> http://zguide.zeromq.org/page:all#A-Load-Balancing-Message-Broker >>>> >>>> Workers are only given work to do if they are free, so you never get >>>> the problem of some workers having a queue of things to process while >>>> others are idle. >>>> >>>> Andy >>>> >>>> >>>> >>>> >>>> >>>> -- >>>> Andy Ballingall >>>> Senior Software Engineer >>>> >>>> The Foundry >>>> 6th Floor, The Communications Building, >>>> 48, Leicester Square, >>>> London, WC2H 7LT, UK >>>> Tel: +44 (0)20 7968 6828 <%2B44%20%280%2920%207968%206828> - Fax: +44 >>>> (0)20 7930 8906 <%2B44%20%280%2920%207930%208906> >>>> Web: http://www.thefoundry.co.uk/ >>>> >>>> The Foundry Visionmongers Ltd. >>>> Registered in England and Wales No: 4642027 >>>> >>>> _______________________________________________ >>>> zeromq-dev mailing list >>>> [email protected] >>>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev >>>> >>>> >>> >>> >>> -- >>> *Nishant Mittal* >>> Director, Product Development >>> *Rosenblatt Securities Inc*. >>> 20 Broad Street >>> New York, NY 10005 >>> >>> Direct: 212-607-3159 >>> Mobile: 646-504-2629 >>> >>> >>> _______________________________________________ >>> zeromq-dev mailing >>> [email protected]http://lists.zeromq.org/mailman/listinfo/zeromq-dev >>> >>> >>> >>> _______________________________________________ >>> zeromq-dev mailing list >>> [email protected] >>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev >>> >>> >> >> >> -- >> *Nishant Mittal* >> Director, Product Development >> *Rosenblatt Securities Inc*. >> 20 Broad Street >> New York, NY 10005 >> >> Direct: 212-607-3159 >> Mobile: 646-504-2629 >> >> _______________________________________________ >> zeromq-dev mailing list >> [email protected] >> http://lists.zeromq.org/mailman/listinfo/zeromq-dev >> >> >
_______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
