conrad@cgl.ucsf.edu (Conrad Huang) (09/25/90)
We have a very strange problem with IP forwarding with a Sun 4/280, for packets from MIPS 3260 and 6280. Out network configuration is MIPS 3260 RISC/os 4.50/4.51 | ...-+--------------+-------------------------+-... | ie1 interface | Sun 4/280 MicroVAX 3200 SunOS 4.0.3 4.3BSD tahoe | le0 interface | ...-+--------------+-... ...-+----------+-... | | SparcStation 1 NeXT Machine SunOS 4.0.3c Mach (Release 1.0) When I run ping from the MIPS to the SparcStation, packets of certain sizes don't get through. For example, (socrates is the MIPS and voltaire is the SparcStation) socrates.28% ping volt 191 2 PING voltaire.mmwb.ucsf.EDU (128.218.21.18): 191 data bytes ----voltaire.mmwb.ucsf.EDU PING Statistics---- 2 packets transmitted, 0 packets received, 100% packet loss However, ping with most other size packets are fine: socrates.31% ping volt 591 2 PING voltaire.mmwb.ucsf.EDU (128.218.21.18): 591 data bytes 599 bytes from 128.218.21.18: icmp_seq=0. time=8. ms 599 bytes from 128.218.21.18: icmp_seq=1. time=7. ms ----voltaire.mmwb.ucsf.EDU PING Statistics---- 2 packets transmitted, 2 packets received, 0% packet loss round-trip (ms) min/avg/max = 7/7/8 Ping'ing from the MIPS to the Sun 4/280 itself works fine at all packet sizes. When I run tcpdump on the SparcStation network, I see that there are many truncated ICMP packets from the 4/280, all short by either 111 or 112 bytes. When I run tcpdump on the main network (MIPS, 4/280, MicroVAX net), the ICMP packets are fine. All this points at Sun networking being buggy. However... the same sized ICMP packets work fine from the MicroVAX to the SparcStation. Ping'ing from the MIPS to the SparcStation, if I manually change the network route to the SparcStation so that it points to the MicroVAX, (so that the packet goes from the MIPS to the MicroVAX to the 4/280 and, finally, to the SparcStations), the one ICMP packet that the MicroVAX sends to the SparcStation gets through. When the route is "fixed" by the ICMP redirect from the MicroVAX, the packets once again stop flowing. If we swap the Ethernet interfaces (le0 and ie1) on the Sun 4/280 so that the le0 interface is connected to the same wire as the MIPS, and the ie1 to the SparcStation, the problem goes away. This is a poor solution, however, as the le0 interface is supposed to be much faster than the ie1, and we want to have the faster interface handling the i/o requests from our diskless clients. So I thought it's a MIPS problem, but ICMP packets from the MIPS to the NeXT machine work fine at all sizes. So then I thought the problem must be hardware, but our other MIPS machine (a 6280 running RISC/os 4.51) has the exact same problem going through another Sun 4/280 gateway. So then I thought I'd post to the net and get some help. HEEELLLLPPPP, Conrad Huang conrad@cgl.ucsf.edu