CD Based NFS Install problem.

[virtech]

So I have my CD based NFS installation working with a Linux VM.

However when I try to replicate this on a physical box the ESX server is not able to mount the NFS directory specified in my .cfg on the network. Instead the language is displayed.

When I manually select NFS Image and point it to my NFS share I can continue the installation.

I'm convinced there is is some sort of network issue here, anyone else had the same problems?

[virtech]

Ok it seems my install is failing becasue the static ip address I have configured in my isolinux.cfg is only coming alive on the network 18seconds after the ESX installation tries to access the source files on the NFS share.

Is there anyway I resolve this?

[virtech]

IBM server using broadcom nics connected to Cisco switch.

Portfast (STP) enabled on port.

Portfast has helped but I still am getting a DHCP time out. I have also tired adding the following in my isolinux.cfg.

ksdevice=eth0 nicdelay=50 linksleep=50 but this hasn't help either.

It would appear the the nic is recycled up to three time during the setup phase before connecting to the NFS share.

Anything else I can try? scratching my head with this one.

[mbrkic]

Are you sure that your NFS server is not doing a reverse lookup on the ESX server's IP and timing out on that (the timeout is typically around 15 secs). Sometimes this is done for logging purposes. Make sure that your NFS server can reverse lookup the ESX IP (the one it is using during installation). If this is the problem you can always add it to the hosts file.

[virtech]

Why would the NFS server need to do a reverse lookup on the ESX servers ip address? I have DHCP and NFS configured on the same box

I am convinced this is a DHCP timeout issue and thats what im seeing when I try and do my build. pump times out

[mbrkic]

Sometimes the logging tries to reverse lookup the IP so it can use the hostname in the log entry. It was just a thought that you can check and eliminate in 2 seconds flat...

I guess it is time to run Ethereal (or some other packet capturing app) on your DHCP servrer, if you can, and see exactly what is happening...

[DFATAnt]

I'm having the same problem and I have checked my DNS and the reverse lookups are there.

I was hoping that the solution could have been that simple, but unfortunately it's not.

Ant

[virtech]

thanks for the hint DNS was one of the first things I checked.

I should be able to look at this more next week, just frustrating when I know it works..But not in this setup

[Gabrie1]

Ever been able to resolve this issue? I'm running into the same here.

Gabrie

[virtech]

Let me guess, are you using broadcom NICS and Cisco switches?

[Gabrie1]

Yep.

I have 2 internal broadcom nics and 4 dual port Intel nics. But the

intel nics are addressed through the broadcom bnx2 driver. (The

internal nics are not used)

[virtech]

Hmm this issue is a real pain, my CD/NFS based build works flawlessly when testing on my laptop using ESX3 in Workstation 6.0 or even in my lab using a netgear GB switch at home.

Have you seen this Red Hat article ?

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=189795

As a test have you put another switch in between your ESX hosts and the cisco switch? I found if I took the cisco switch out of the equation my problem went away.

[Gabrie1]

This is going to be a real problem. Gonna make a support call with VMware but I'm affraid they wont be able to help.

Gabrie

[virtech]

I don't like your chances, this appears to be be Red Hat issue but I guess we need both parties to start talking to one another.

On the off chance you make any progress, please post back

[Texiwill]

Hello,

Could you list out the following:

OS and Version of NFS server and of NFS.

OS and Version of DHCP server and of DHCP.

Does the same problem occur if you switch to a HTTP based install?

I have used both NFS and HTTP with great success with broadcom NICs and Cisco routers.

Best regards,

Edward

[Gabrie1]

Hi

The PXE boot server is the UDA 1.4 from Mike Laverick. Its running apache http. I will have to look what DHCP version there is in this image.

A network engineer played with the port speed and now we have all vmnic0 nics (on all 4 esx hosts) running at 100Mb Half Duplex and it works \!!! They run the installation without problems.

Still investigating !!

Gabrie

[Texiwill]

Hello,

So you are using HTTPD instead of NFS to transfer the files?

Best regards,

Edward

[Gabrie1]

Yeah and we're proud of it

(Sorry, has been a long day )

[Texiwill]

Hello,

I do that as well. I personally like it better. I do not use UDA however, I built a Fedora Core 6 PXEboot server. Never a problem. Granted my DHCP server is on a completely separate box.

Best regards,

Edward

[Goof]

The problem with broadcom nics is a DHCP problem. When I add the "ip=" option to the kernel commandline the kickstartfile can be loaded and parsed.

I load the kickstartfile from http and the iso images from NFS.

To debug the problems the 'Alt-F3' and 'Alt-F4' screens are very usefull.