Your code in 2.6.33 kernel introduced a new system call called
recvmmsg(), a syscall for batching recvmsg() calls. Would you like to
share the set of test cases you used to test this new code with the LTP
On Wed, 2010-05-12 at 18:20 -0300, Arnaldo Carvalho de Melo wrote:
> Em Wed, May 12, 2010 at 11:10:23PM +0530, Subrata Modak escreveu:
> > Hello Arnaldo,
> > Your code in 2.6.33 kernel introduced a new system call called
> > recvmmsg(), a syscall for batching recvmsg() calls. Would you like to
> > share the set of test cases you used to test this new code with the LTP
> > community ?
> Forgive me my ignorance about the current state of the LTP
> infrastructure, but does it support requiring multiple clients and a
> server to test networking calls? Or is everything done over loopback?
Sorry for the very late reply. No, we also have something called LTP
network tests, which require a separate server and client, and is
designed to burn the network wires and not the loopback interface.
Your test seems interesting. I will like to integrate the code later in
our networking tests when i find time. If there is a more active
volunteer, he is welcome to do it before me ;-)
> The tests I did were basically:
> On several client machines:
> use netcat to continuously send datagrams to a destination machine on
> a specified port (5001, the ttcp one is used as a default)
> On the server:
> run the attached testcase, then I was basically visually looking that
> every batch received packets from multiple clients.
> I can envision a test where from the machine chosen to be the server a
> script like:
> ./recvmmsg &
> for client in $CLIENTS ; do
> ssh client ./run-netcat-script.sh $SERVER
> # --- end ---
> The attached recvmmsg would somehow get the list of clients, check that
> they are the sendeds in datagrams in the same batch, and if so, woud
> emit a PASSED as output.
> Another question about the LTP infrastructure: Do it already supports
> collecting measurements in databases, associated to the kernel, libc,
> etc, versions, so that this can be used to generate plottings of metrics
> over time/kernel versions?
> - Arnaldo