RTEMS networking performance issue
Michael J. Siers
mikes at poliac.com
Mon Jul 22 14:12:11 UTC 2002
I just noticed a post of a patch to NetBSD and OpenBSD for a TCP Socket
This patch effectively makes the performance of TCP buffering go from
O(N) to O(C),
independent of the size of the list to traverse. This NetBSD developer
has also made
some patches for doing zero copy TCP and UDP (this patch seems to be a
I do not know how different the NetBSD/OpenBSD network stack is from the
RTEMS/FreeBSD stack but it might be interesting to see if these patches
made to work in RTEMS. (I just wish I had some time to play with it!).
Check out this site for more information (it has links to the
appropriate patch files and
jvegiraju at netscape.net wrote:
> I am working on porting RTEMS networking code to ARM9 processor based system.
>The ttcp benchmark test gives me 12Mbps data throughput for TCP test on a 100Mbps switched connection(The RTEMS box is directly connected a linux box).
>Here are some details about the test setup.
>I am using RTEMS 4.5 release and ttcp app from netdemos-4.5.
>The task priorities are,
> netdaemon task - 100, enet_drv_tx_daemon - 75, enet_drv_rx_daemon - 75,
> ttcp app task - 150.
>The ethernet driver has a TX queue. The queue size is never going beyond 3 packets(while transmitting from RTEMS side to linux side). I see a TCP ACK from linux box after every two packets that it receives.
>The CPU utilization is 100% during the test. It looks like the networkdaemon task is spending too much time in either packet processing or busy waiting for an event to happen.
>I am using in_cksum routine optimized for ARM.
>I am wondering where the bottleneck is.
>I would appreciate your insight into this problem.
>Your favorite stores, helpful shopping tools and great gift ideas. Experience the convenience of buying online with Shop at Netscape! http://shopnow.netscape.com/
>Get your own FREE, personal Netscape Mail account today at http://webmail.netscape.com/
More information about the users