I have been working on an automated build based around Kickstart, and thought it may help some people to note a slightly off topic) issue that may be encountered by others. My configuration involves 2 x IBM x330 systems (with onboard Intel Ethernet Pro 100 dual port NIC's) connected via a Cisco 2900XL Catalyst switch. One IBMx330 is the server (for everything), and one is the install target. The objective is an entirely media-less boot mechanism. This is working using PXELinux communicating with DHCP, the client obtains the kernel and ram disk via TFTP, and then picks up the ks.cfg over NFS, which in turn issues a static IP and installs over NFS. The reason for this diatribe is that I experienced a problem whereby, if I interrupted the boot process (Ctrl-S) when it reached the /sbin/loader, and then released it (Ctrl-Q) the install worked. If I didn't interupt it the install would hang. In the tcpdump I could see the client issuing DHCPDISCOVER requests, and the server responding with DHCPOFFER's. But, the client would not accept the offers and therefore could not get the ks.cfg. This is mentioned in the DHCP FAQ as a possible routing issue, but there is another possible cause. Turns out that the ports on the Cisco were set to auto-negotiate, with spanning-tree enabled. After setting the ports to fixed 100MB Full Duplex with no spanning-tree, all works fine. Apologies for the purists that will tell me this is a network/DHCP issue, but I thought it might be of interest to some of this group as well. Regards, Pete. ---------------------------------------- Send 10 free text messages from your PC to your friends' mobile everyday! Plus Vizzavi offers the best up-to-the-minute news, sport, entertainment and more on web, WAP and mobile. http://www.vizzavi.co.uk ----------------------------------------
