[zeromq-dev] P/Invoke-based CLR access to zeroMQ

[Martin Sustrik]

Dirk,

Is that on a single box via loopback interface? The performance on 
loopback tends to be quite unstable. If you want to get some real 
numbers I would advise using two separate boxes. Also number of messages 
in the test is important. The higher the number, the more reliable the 
result. If you use something like 1000 messages, the result would vary 
by an order of magnitude. We are normally testing with batches ~10M 
messages.

Martin

dok at dok-net.net wrote:
> Well, I spoke to early, my latency test missed the message aggregation 
> feature of 0MQ, sending each 1 byte message on a synchronous roundtrip.
> 
> If I just measure how many 1 byte messages can be transmitted per second 
> in one direction, the current C# wrapper achieves > 188000 messages/s 
> (5.3microseconds/message). Using this "oneway" test for throughput 
> testing, I get 78MiB/s. (all on an Athlon 64 X2 5000+).
> 
> -Dirk
>