Andrew Boyd wrote:
>
> You didn't say what cpu and network hardware you were running on.
>
> We have done these sorts of tests before.  IIRC between speedy x86
> and gige I think qnet turned around a send/receive/reply in 200
> microseconds.  This was with a direct point-to-point link - no switch,
> no other traffic.

   The middleware Zero MQ does the same send/receive/reply cycle in 86us 
... 10 bytes per message.

   --Armin



>
> --
> aboyd
>
> ________________________________
>
> From: John Pruitt [mailto:community-noreply@qnx.com]
> Sent: Wed 9/29/2010 1:18 PM
> To: general-networking
> Subject: QNET performance
>
>
>
> I am looking for information on QNET performance. I have a simple resource manager that simply lights up an led on my 
target board. I have constructed a test program that
>
> 1. opens the file
> 2. reads the number of ClockCycles,
> 3. in a for loop, does an unbuffered write to the open file to turn the led on or off
> 4. reads the number of ClockCycles
>
> If this is done locally, the time per write is measured at around 100 microseconds. I assume that most of this time is
 the time needed to get from the user program and into the resource manager.
>
> If this is done remotely over QNET, the time per write is measured at around 7000 microseconds or about 7 milliseconds
. Does this seem like a reasonable time?
>
> Are there any performance measurements taken regarding QNET?
>
> Is there anything like the Nagle algorithm for TCP/IP that comes into play with QNET? Are there any parameters for 
QNET that allow tuning for performance?
>
> Will this number be real sensitive to whether the two machines are on a private sub-net?
>
> Thanks.
>
>
>
> _______________________________________________
>
> General
> http://community.qnx.com/sf/go/post69097
>
>
>
>
>
>
>
> _______________________________________________
>
> General
> http://community.qnx.com/sf/go/post69114
>
>