Existing PXE not working with Flex-10

[clint0r]

I'm having problems getting PXE to work on anything connected to my flex-10 modules and wondering if someone can help shed some light on why I'm having issues. PXE server is Altiris DS 6.9 SP1 DHCP server is on a seperate server, IP helpers are in place for both PXE and DHCP. Firmware is up to date as per the 8.40 ISO Both Flex-10 modules are running firmware 2.01 Blade Configuration: Bay 1 - Flex 10 (1Gb copper uplink to Cisco 3560, vlan trunked) Bay 2 - Flex 10 (1Gb copper uplink to Cisco 3560, vlan trunked) Bay 3 - Cisco 3020 (nfs vlan) Bay 4 - Cisco 3020 (trunked vlans for vm network, identical to bay1/2 trunks) I've tried ruling out PXE and DHCP by: - creating vm's that use the 3020 in Bay4. They can pxe boot to our altiris server successfully. - powering on new 360 G5's connected to the same 3560. these servers can pxe boot on all vlans. Any physical blade on the flex 10 can get DHCP on any vlan if it's running esx or windows, but on the pxe end I get different errors for different types of blades. BL460 G1: PXE-E61: Media test Failure, check cable PXE-M0F: Existing Broadcom PXE ROM. BL495 G5: Waits on trying to get a DHCP address and eventually times out. I've tried changing PXE from Bios to Enabled in Virtual Connect Manger with no success. What am I missing? Cheers Clint

[Neal Bowman]

Clint, I've not done anything with the Flex10 modules yet, but in your network configuration, are you setting the VLAN tunneling or VLAN Mapping mode? If mapping, have you defined the default VLAN? Neal

[clint0r]

Hi Neal, Virtual Connect is configured for VLAN mapping. I have not set a native VLAN but both embedded nics are set to the same vlan and can get ip addresses through DHCP if I bypass PXE and boot into an operating system. I doubt this matters, but I have 20 ethernet networks defined in a single uplink. Clint

[Neal Bowman]

Hey Clint, One way that I got around this was to have a separate uplink network configured as my "ServerBuild" port. On the Cisco side, it is configured as an access port, and not a trunk port. I changed the networking in the server profile from ServerBuild to the other trunk port after the build is completed. Changing the NIC assignment in the profile does not require powering off the server, but the IP address will normally change. Neal

[clint0r]

Hi Neal, I see how your workaround would work. Unfortunately, I am demo'ing these modules and do not have any extra SFP's to create a new uplink. Does it matter where my copper uplinks are plugged into? Currently I have them in Port x1 and am wondering if they should be in x2-6 or the shared uplinks in x7/8. Cheers cp [Updated on 4/1/2009 12:42 PM]

[Neal Bowman]

Clint, I just use the RJ45 copper connections for the build network, not fiber. Using the VC profile, you can assign to any NIC that can PXE boot. Neal

[Neal Bowman]

Hey Clint, I just ran across this advisory from HP, sounds vaguely familiar. :-) http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=us&objectID=c01683477 Neal

[clint0r]

Hi Neal, Thanks for the link - that fixed the PXE issues I was having on the BL460c by changing the link speed from Auto to 1000 full. I'm still having problems getting the 495 to pxe boot (pxe-e51) but I am guessing it's a firmware issue as well. Thanks! Clint

[Calvin Staples]

Just wondering... Are you using "portfast trunk" (for trunks) or "portfast" on your Cisco ports? That may well take care of your timeout issues and in the case of a link failure will allow your servers to reconnect in seconds instead of 30 seconds or more. We had a problem that sounds much like yours. After enabling the portfast option have had no more issues.

[clint0r]

Thanks for the suggestion. I had portfast enabled on all our trunks which doesn't seem to have helped. I have ordered a couple more copper modules to do some non-trunked testing.