Evert Lammerts wrote: >> Imre Paniti wrote: >> >>> Hi, >>> >>> >>> I'm getting a lot of ENOBUFS while repeatingly sending over 10 bytes of >>> data data from a fast rt thread (100 packets per second). I guess this >>> makes >> I wouldn't consider 100 Hz as fast. Is this the correct rate? Or is the >> receiver overloaded and does some MAC layer overflow signalling? > > The rate is correct, and the reciever is probably not overloaded. > >>> sense - no available buffers. But how does this work internally, what >>> happens with a que on a socket (we're using UDP, might a raw socket be >>> better?) when writing to it fast? I know we can increase the buffer >>> size, but that would only mean it fills up less fast but still fills up. >>> So on >>> the practical side of things, what can we do about this -105? >>> >>> We're using: >>> RTAI-3.3-CV >>> RTnet 0.9.3 >>> >> With or without RTmac? If /wo RTmac: which NIC driver? Switched Ethernet? > > This is lsmod after a simple start-up routine of RTnet: > > emc2-2:~# lsmod > Module Size Used by > tdma 14852 1 > rtmac 8976 1 tdma > rtcfg 37168 0 > rt_loopback 2564 1 > rt_8139too 11776 2 > rtpacket 4352 0 > rtipv4 21536 1 rtcfg > rtnet 29292 7 > tdma,rtmac,rtcfg,rt_loopback,rt_8139too,rtpacket,rtipv4 > rtai_rtdm 16676 6 tdma,rtcfg,rt_8139too,rtpacket,rtipv4,rtnet > rtai_sem 15232 4 tdma,rtcfg,rtnet,rtai_rtdm > rtai_up 71736 8 > tdma,rtcfg,rt_loopback,rt_8139too,rtipv4,rtnet,rtai_rtdm,rtai_sem > rtai_hal 29176 6 > rtmac,rt_8139too,rtnet,rtai_rtdm,rtai_sem,rtai_up > > This means RTmac is used (right?). As you can see we're using a realtek > 8139 NIC. > >> >> There are at least two error sources involved when sending out packets: >> first the per-socket pool of rtskbs that is charged when the outgoing >> frames are to be build. Later the NIC may report an overflow (EAGAIN) of >> its own hardware TX queue - unless you are using RTmac as flow control >> manager. This is true for both UDP and packet sockets. > > So I guess this means the error comes from rtskbs, because we're using > RTmac? If so, then what does that mean?
This means now that you do not provide enough bandwidth to your transmitter, thus it enqueues more packets then there are transmitted and runs out of buffers sooner or later. How does your TDMA configuration look like (see also /proc/rtnet/rtmac/tdma*? And how many other senders are sharing the transmission time slot(s) of the particular box? > > It might be useful for you to take a look at my code. However, this is for > a module for CNC software (Enhanced Machine Controller 2, EMC2) and it > uses a custom RT API in combination with the RTnet and RTDM API, which > makes it a little harder to read. Apart from that, in the actual functions > that are exported as RT tasks, nothing really strange happens - a socket > is created during module start-up, the RT tasks only either create packets > and send them or recieve packets and disect them. In fact, your problem should be independent of the actual application code. Jan
signature.asc
Description: OpenPGP digital signature
------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________ RTnet-users mailing list RTnet-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/rtnet-users
