I'm seeing a strange error behavior on a 6509/SUP720. The net effect is that uploads to connected ISPs run slow. Very slow. Like, DSL slow, not the gigabit speeds of the actual connections. But tracking it down, the symptoms are weirder.
Namely:
We're seeing nasty packet loss, but the loss rate is a factor of hop count to the destination and packet rate.
Directly connected devices routing over the 6509 (2 hops, by MTR's count) can run hundreds of pings a second before they start to drop.
A destination 3 hops away may be rock solid at hundreds of packets a second or may start to lose packets at less than 100pps.
By five hops, packet loss is guaranteed even as low as 30pps.
By seven hops, packet loss at 10pps is common.
Packet loss scales with distance and pps count, so if 30pps causes 30% loss for a given destination, 100pps causes 50% loss for the same destination and so on. Different destinations start losing packets at different hop counts and speeds, but it's always a steady increase in loss when pinging each hop further along a path to given destination.
Strangeness:
MTR reports the loss at the first hop, with no added loss per hop. So 3 hops may generate 0% loss, but 5 hops generates 30% loss at the first hop and all after.
Anybody seen anything like this before?
