Assuming you’re using TCP, the message format is identical between Windows
and Linux. (If you’re using IPC, then that’s not true, as Windows uses
Named Pipes which are quite different from UNIX domain sockets used on
Linux.)
Now, there are lots of different things under the hood — scheduling, etc.
which can cause differences in behavior, usually as a result of faulty code
(in the library or in the application.)
One thing that you should be aware of — are you using x86 CPUs in both
cases? If the processor architecture is different you may find that there
endianness considerations that you’ve not handled in your application
correctly. (For example if you have an ARM CPU running in big-endian mode,
and are just writing integers without handling byte order conversions in
your application payload.)
I’d have to see some code to debug this further. It’s always best when
submitting code to try to pare it down to the minimal amount of C code to
demonstrate the problem, if you can.
Thanks.
- Garrett
On Mon, Apr 23, 2018 at 7:42 PM ^^ <2mailme@xxxxxxx> wrote:
Dear Sir,
Sorry for disturbing you.
I have a problem which is confusing me.
I develop a program which has a producer, 2 workers, and a collector.
Producer read file and send raw binary data frame by frame to workers,
workers process(just add header and rearrange line of raw data) and send it
to collector.
1) The pattern I used is PUSH – PULL:
producer works as server, push data;
workers work as client, pull data, after processing then push data to
collector;
Collector works as server, pull data from workers.
First time, I deployed producer, workers and collector to a Linux server,
I used different port to separate 2 workers, everything works good.
For testing purpose, I moved a worker to a Windows server, then headache
thing happens, the data processed by worker on windows is corrupted.
Then I closed the worker on Linux, only let the worker on Windows alive,
all data received by collector is corrupt.
I tested multiple time, on Linux, no matter how many workers there are,
the data outputted by collector is correct, but if data transferred to
Windows then return, only first few bytes are correct, others are corrputed.
All nn_send and nn_recv use NN_MSG flag, and the data size is about 700KB.
Linux and Windows worker’s source code is exactly same.
The nanomsg version both on Windows and Linux is 1.1.2
I am wondering is there differences between Linux and Windows message
format or messaging mechanism?
Or just I missed something important.
If necessary, I will send you my code.
Hope to have your help.
Thank you.
