NetMQ now supports curve. Anyway I suggest to set both Send and Receive Highwatermarks, especially when using curve. Make sure to set it on edge sockets and proxy. I don't suggest to use zero but to calculate the number according to your system tolerance. Also, you might want to increase zeromq IO threads (default is one) as encryption is CPU heavy.
On Mon, Apr 27, 2020, 21:13 up201503995 via zeromq-dev < [email protected]> wrote: > I am doing some performance tests on ZeroMQ to compare it with others > like RabbitMQ and ActiveMQ. > > In my broadcast tests and to avoid "The Dynamic Discovery Problem" as > referred by ZeroMQ documentation I have used a proxy. In my scenario, I > am using 50 concurrent publishers each one sending 500 messages with 1ms > delay between sends. Each message is then read by 50 subscribers. And as > I said I am losing messages, each of the subscribers should receive a > total of 25000 messages and they are each receiving between 5000 and > 10000 messages only. > > I am using Windows and C# .Net client clrzmq4 (4.1.0.31). > Code to reproduce the issue is available at > https://pastebin.com/SEqEhSL1 > > > I have already tried some solutions that I found on some posts: > > I have set linger to TimeSpan.MaxValue > I have set ReceiveHighWatermark to 0 (as it is presented as > infinite, but I have tried also Int32.MaxValue) > I have set checked for slow start receivers, I made receivers start > some seconds before publishers > I had to make sure that no garbage collection is made to the socket > instances (linger should do it but to make sure) > I have a similar scenario (with similar logic) using NetMQ and it > works fine. The other scenario does not use security though and this one > does (and that's also the reason why I use clrzmq in this one because I > need client authentication with certificates that is not yet possible on > NetMQ). > I have also tested this scenario with clrzmq but with no security > and it worked > > Does someone know another thing to check? Or has it happened to anyone > before? > > Thanks, > > Davide Costa > _______________________________________________ > zeromq-dev mailing list > [email protected] > https://lists.zeromq.org/mailman/listinfo/zeromq-dev >
_______________________________________________ zeromq-dev mailing list [email protected] https://lists.zeromq.org/mailman/listinfo/zeromq-dev
