I have a Teradata Machine, using MP-RAS Unix, with a 1000 Intel Ethernet card and a Cisco switch.
If I configure the ethernet card and the switch to auto, so they negotiate to 1000, or configure the ethernet card and switch manually to 1000Full or 100Full, the velocity is very very low.
Only works right If I configure then to 100Half manually.
Anybody know Why this happen and How can I fix It?
Thanks.
What type of cicso switch? What version of code on the cisco switch?
Sounds like a duplex mismatch - one end is half and the other full. I know you've already checked, but do make sure that both ends are set to full.
Cisco have a good troubleshooting guide on their website (Networking, Cloud, and Cybersecurity Solutions - Cisco). Your symptoms align well with their description of duplex mismatch.
However, note also that the section on "NIC Compatibility" has Intel Pro/1000 listed as a known problem, for which the "latest drivers" are required, so maybe that's worth a try.
There is a note on the Teradata Forum (albeit four years old) that suggests turning off "in-line power" from the switch. Re: Slow Responding State - looking for bottleneck [From Anomy Anom: Fri, 18 Jun 2004 @ 17:07 GMT]
Once the basic connectivity is sorted don't forget to enable jumbo frames on the switch and set the NICs mtu to 9000 and enable "full-size frames". You'll probably also need to increase your TCP send and receive buffer sizes (to at least 1000Mbit times your ping time in seconds) in order to get the best throughput. All these parameters can be seen using ifconfig.
Finally, if your Teradata/MP-RAS machines were not shipped to you with gigabit NICs (i.e. you've upgraded some old machines) then you may need to increase the MAXETHERPACKETSIZE kernel variable and recompile/reboot in order to get the full 9000 packet size. Use idtune -g MAXETHERPACKETSIZE to see what it's set to.
HTH.