Improbable packet loss to nv, bug?

David H. Brown (dhbrown@jeeves.nsi.nasa.gov)
Wed, 8 Feb 1995 14:22:35 -0500


CU-SeeMe reports (seemingly) erroneous packet loss rates to an nv client
via multicast, forcing sender rate to back off to minimum.

Config:
Machine 1 - Sparc10, SunOS4.1.3 w/MCAST
running Reflector 3.0b3 w/nv and vat multicast .conf

Machine 2 - SparcIPC, SunOS4.1.3 w/MCAST
running nv v3.3 beta, vat v3.4

Machine 3 - PPC8100/80AV, MacOS7.1.2
running CU-SeeMe (0.8b1, 0.8b2)

Symptoms:

I have successfully configured the above to pass CU-SeeMe format video
between machines 2 and 3 above via multicast, (I will try to htmlize the
procedure for posting on appropriate sites). I have noticed an extreme
performance problem that often though not always occurs. Machine 3's cap
gets driven from its local defined value (80kbps) to 10kbps. When I look
at machine 2's packet stats, (as viewed from machine 3), I see packet
loss rates in the range of 10's to 100's of thousands out of a total
transmitted on the order of thousands. There is no way this is
correct, since not enough time has elapsed for me to send that many
packets. Loss rates quickly jump to the high 90%'s. This of course causes
machine 3 to back off its transmit rate. Packet loss almost always
jumps in blocks of 28416. I have also seen negative values for packet loss
(register overflow?), resulting in greater than 100% packet loss
reported. I can get around this problem by forcing
machine 3 to ignore the choke back (i.e. set min=max=80kbps), which
results in good performance. I also notice nv on machine 2 reports high
packet loss (~8-15%) when machine 3 is choked back to 10kbps, but reports
better stats when machine 3 is forced to transmit at 80kbps (0-2%).

I have not seen these problems when running in Mac/PC client only mode.

Is this a known bug? Is it a new bug? Is it me? Please advise.

Thanks for a great product.

Regards,
Dave Brown