TCP errors

Hi all,

I've been running some test programs, telnet, and an application that uses ONC RPC over TCP and I'm seeing unstable throughput, telnet sometimes hangs, and in Wireshark on the host and tcpdump on the device I see various TCP errors such as retransmissions, duplicate ACKs and out-of-order. These errors show up the most when the device is connected directly to my PC or through a dedicated gigabit switch. Running commands via telnet in this configuration is sometimes unusable as the connection hangs. They show up less often when the device is connected to a 10/100 switch on a network with lots of other PCs on it, but when they do, they reduce the bandwidth dramatically for several seconds.

I've tried both a Windows and Linux host, both with Gigabit Ethernet.

I've been playing around with all the Linux TCP options (/proc/sys/net/ipv4/tcp_*, SO_SNDBUF, SO_TCPBUF, TCP_NODELAY) and nothing seems to make TCP communication stable, although using 8KiB socket buffers seems to be better than just about any other size. I've checked duplex and speed settings and tried setting them manually with autonegotiate turned off.

I am using the Altera TSE with the SLS driver and a DP83865 PHY. I also tried briefly with the NEEK, which is almost identical except for the PHY, with the neek_web_server_mmu design and also saw the problem.

This is the closest thing I've seen to describing the problem, though it could be a completely different thing: https://linuxlink.timesys.com/forum/2037

Anybody else seen problems with the TSE and TCP? Any other Linux tweaks to look at?