Tuesday, December 29. 2009
More robust network booting Posted by Michael Prokop in general at 22:00
Summary: Grml gets more robust support for network booting.
Background information: A customer uses Grml for the deployment process of their systems. Grml isn't just used for the deployment but also as part of the netboot environment in the High-Availability (HA) setup running on IBM BladeCenter systems. The netboot setup is used for hardware inventory and to be able to replace bladesystems without the need for any manual configuration (implementing features of IBM BladeCenter Open Fabric Manager using Debian/Linux). As being part of a HA setup the netboot setup should work no matter which server is unavailable or which network connection is broken. The software stack is redundant thanks to DRBD, Heartbeat& CO while the hardware stack itself is redundant due to the bladeserver infrastructure.
Problem description: If you have more than one network card (NIC) inside your system you might encounter problems with network booting as soon as the first/preferred NIC doesn't work. This is a common problem in netboot environments, usually solved by crude hacks and workarounds.
Solution: The last Grml release (2009.10) already invented the ethdevice= bootoption which allows you to specify a specific NIC for booting. I just extended ethdevice and its surrounding code so it is possible to specify multiple devices at once that should be configured. If you don't have any specific configuration all present NICs will be used for configuration via DHCP automatically. The resulting code is quite tricky because ipconfig of klibc-utils might fail in several situations. I just uploaded live-initramfs (1.157.4-1grml.01) to the grml-testing repository, the daily ISOs will provide the feature within the next few days as well.
Available bootoptions and behaviour:
Screenshot of netboot in action:
« previous page (Page 1 of 1, totaling 1 entries) next page »