On 11/4/05, Grant Grundler <[EMAIL PROTECTED]> wrote: > On Fri, Nov 04, 2005 at 10:19:52AM -0800, Grant Grundler wrote: > ... > > o The comments in "Recv operations" talk about "backpressure". > > Is this another way of saying the driver should drop packets once > > the "fairness threshold" is exceeded? > > Ranjot's slideset answered this question (I think): > | o Slow receiver ports are stalled at sender side > | - combination of activity (LRU) and memory utilization used > | to detect slow receivers > | - sendmsg() to stalled destination port returns > | EWOULDBLOCK, application can retry > | - recvmsg() on a stalled port un-stalls it > > I'm having trouble reconciling previous "connectionless" and > "transperent to user space" comments this this slide. > Especially the "EWOULDBLOCK" return code. > > If a reciever can cause a sender to stall, it implies the packets > will get dropped on the send side. This is a subtle change > in behavior that I don't think any UDP application can assume. > But I'm no networking protocol expert...
When the sender is stalled, the driver will backpressure the application.. no packets will be dropped. Since a UDP application assumes the underlying transport is unrealiable it should not have any problems running on RDS. On getting EWOUDBLOCK it will simply retry. > > thanks, > grant > _______________________________________________ > openib-general mailing list > [email protected] > http://openib.org/mailman/listinfo/openib-general > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general > _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
