Teresa Noviello wrote: > On 11/3/06, Jan Kiszka <[EMAIL PROTECTED]> wrote: >> >> Teresa Noviello wrote: >> > Hi. I have a master, 10.0.0.4 and a slave 10.0.0.5. >> > >> > Starting rtnet with "rtnet start" on the slave 10.0.0.5, i have seen >> that >> > time difference with the master is about 19microsec (min 19, max 20). >> >> You mean the calibration result in the kernel log? That is the time it >> takes (averaged) from the master writing its current time into a frame >> to the slave comparing that value with its local time after receiving >> the related frame. >> >> > >> > With ethereal i do LAN sniffing, running ethereal on rteth0 on master, >> > 10.0.0.4. >> >> If you sniff on one node participating in the network, you will face an >> inherent offset between timestamps reported for outgoing and incoming >> packets. The former are stamped *before* registering with the hardware, >> the latter after reception (including potential IRQ jitters). >> >> > >> > With ethereal i see TDMA Synchronisation, Heartbeats and so on... >> >> (Hope you read in the docs about the different protocols behind TDMA >> sync and RTcfg heartbeat. In this context, those heartbeat frames should >> be considered as an arbitrary payload for TDMA.) >> >> > >> > Just to understand about timing: i see the tdma Synchronisation at time >> > 0.608994 and the following heartbeat of the slave 10.0.0.5 to the >> master >> at >> > time 0.609136. The difference is 142 microsec, so the heartbeat of >> slave >> > 10.0.0.5 arrives at master after 142 microsec since the >> synchronisation. >> > Now >> > considering the tdma.conf i have written gives slot >> > 0 0 to master >> > and slot >> > 0 120 to slave >> > (the cycle is 1000) >> > >> > i can say that the heartbeat start at 120microsec after >> synchronisation, >> so >> > the real time the heartbeat (60 bytes) hold the cable is >> > 142 - 120 = 42 microsecs. >> >> (guess you mean 22 us) > > > > Yes, excuse me, 22 us > > >> >> > Now, i want to understand: i do have to add the delay (19 microsecs) of >> the >> > slave 10.0.0.5 to this time to calculate the real time the heartbeat >> holds >> > the cable? What is the relation of this delay on the times i do snif >> with >> > ethereal, on master? >> >> In the unloaded, almost jitter-free case you actually see that packet >> travelling delay here again, but now the other way around >> (slave-to-master). Additionally, you face the wake-up latency of a timed >> task on the slave, because the calibration doesn't account for this. >> >> Jan > > > > So, the time my heartbeat hold the cable standing on ethereal (and > considering jitters==0) Is actually 22us? Or is it 22+19? >
Hmm, not sure if I got what you mean: The actual transmission of a frame is only one part of those 20 us (7 us or so). The other parts relate to interaction with the NICs (command and data transfer) and the interrupt latency on the receiver side. 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
