I have an intresting issue with my network, I have a Sunfire V100 with dual nics that will not get any errors if I download data from the internet, but will get tons of FIFO (overruns), errors and frame errors when I copy data on my LAN

Everything is connected to a switch, managed all duplex settings checkout all speed settings checkout, The sunfire v100 has 2 10/100 nics on it, everything else on the LAN uses gigabit.

So why am I getting so many errors going from 1G to 100M, but doing 100M to 1G works? (my router is gigabit)

The odd part is there only on the RX side of things, the TX side has no errors on the internet or the LAN...

The only thing I can think of is maybe the Davicom NIC chips on the V100 have very small buffers?

Should I be concerned? I hope to use the v100 as a firewall as it is a 1RU and only uses 45watts on load!

Any input would be helpful,
thanks

Based on your description seems like the Gigabit router doesn't recognize the fact that your appliance has maximum 100 Megabit speed. Find a way to configure your router to know that it is connected with lower speed equipment and that will likely resolve your problems. The reason why it doesn't occur in the reverse direction is that the router is able to keep up with the data rate. Further likely is that in 1 Gig to 100 Meg direction, those extra large packets are being sent and the 100M side doesn't know about those at all, but in the 100M to 1Gig direction, there are none of those packets, and nor would it matter because in that case the receiver is capable of accepting those.

Ummm. Sorry but the above post reveals some misunderstandings about ethernet. If you connect a FastEthernet Port to a Gigabit Ethernet port, one of two things will happen. Autonegotiation will fail and the link won't function AT ALL - the transmitter is clocking data out at one speed and the receiver is reading it at another. If autonegotiation is successful both ends will run at 100Mbps. Therefore the switch cannot send more data than 100Mbps.

Any packet drops due to bottlenecking in traffic heading towards the server would occur in the switches egress queue not at the server receiving end.

It is possible that the network stack on the server is for some reason not servicing interrupts from the NIC card, which would result in buffer overruns (packet drops) but this will not produce Frame Errors.

Have you definitely checked the Duplex state, not just checking the config, but checking the actual port status? I have seen Sun machines configured as full duplex that quite clearly weren't. ethtool will show you the negotiated rate.

Otherwise this is probably more like a hardware problem or a driver problem. You could try disabling or enabling offloading features like checksum offloading, rx coalesce buffers etc, this sometimes offers a workaround to driver problems.

Hey Pete,

The V100 uses tulip (its a Davicom chipset) I do force the duplex settings on both switch and v100 to 100Mib on the tulip driver this can be done with options=5,5 full_duplex=1,1
Running ethtool tells me it's running in 100Mib mode, but that's all it will allow me to see, the tulip driver has no support for setting most of these options. Also note that flow control on these devices is also not supported.

I have checked both speed and duplex settings on both the v100 and the switch.
When I wget a file on the v100 I get about a 5.6MB/s transfer rate (this is going to a switch then to a gigabit router, then out to the internet) I get no errors
However, If I now try to copy that file from the V100 to another box I get the file, with a much slower transfer speed and frame errors, dropped packets and overruns.

How is that even possible, the router I have is Gigabit, if it were a clocking issue then how can the router fire it's packets at the v100 without any error at all? Yet the box on the same switch gets all these errors when talking directly to the v100?

For purpose of complete information:
OS: Debian 7.0.1sparc64
NIC driver: tulip

Found the problem, had nothing to do with the NIC's, I had some bad ECC memory on my V100 that was the problem.

Data transfer errors but not kernel or app crashes? Interesting. Well spotted :-)

I normally wouldn't have caught it, what I did was enter LOM (lights out management) #.
set the bootmode to diag then reset.

When it went through it's test's, it spotted a few ECC errors, after removing those effected DIMMS, booted it back to normal
and my connection was far more stable. after sending 40GiB through a 10/100 interface I hit 2 overruns and no errors.