Monday, April 6, 2009

The table cloth trick

The machine I do my simulation work has 4 x 320GB drives arranged in a RAID10 array for the host OS, a smallish RAID0 array for swap space, and a RAID5 array for my data including the two VMs, one a Windows XP for writing the other a Linux VM for coding. I have been a Fedoa user since Core 1. The host OS was Fedora 8 and the guest Fedora 9. However, the system was crashing several times a day - the mouse and keyboard would freeze in the guest and the only way out was a hard reset.

Since Fedora is not one of VMWare's supported OSs, I decided I needed to try Ubuntu, which is. The question was: could I pull Fedora out from under all this without disturbing the raid array with my data and my VM disks. Ths process was complicated by the fact that the Ubuntu installer does not have RAID10 support.

So I thought back to my early days as an IBM SE and wondered what would I have done with VM (back then VM was an IBMoperating system product - perhaps it still is)? The beauty of VM was that you could run any operating under it to test stuff out before going live on the bare iron. With this in mind, I set out to create a virtual replica of the physical system under VMWare.

I set up a test Fedora guest with four virtual disks and set up raid partitions to match the host configuration. Some surfing turned up a procedure for breaking into the Ubuntu installation and starting the raid system and then mapping the drives before copying the files into the new file system and completing the install. This is the kind of process that often goes off the rails half way though and that's the point at which a hasty and often irreversible decision adds another day and considerabel hand-wringing.

I practiced this install three times from start to finish using this test Fedora guest VM, each time deleting the newly created Ubuntu guest and replacing it with a copy of the original Fedora one. When I felt fairly comfortable, I did the same thing on the bare iron - and voila, it's working.

For me this kind of major upheaval close to the hardware would normally have been the cause of a huge amount of grief; testing and practicing the procedure in advance with VMware was invaluable. (I feel rather like the magician who's just pulled the tablecloth out from under a fully set table of dishes, glasses and cutlery).

Now all that's left is to see whether Fedora was indeed the cause of the problem.