Hi, On Thu, 31 Jan 2002, Calyth wrote:
> Although I don't have the experience, I'd advise against using a soft-RAID > root > partition because if you don't have the right kernel for recovery.... you > could > be in pretty big trouble. I've been running potato with the root fs on RAID-5 on several servers. You could be in trouble if you had no rescue floppy. But I found it was a cinch to replace the standard kernel image on the potato rescue floppy with an image of the RAID kernel I'm using (2.2.19 with the RAID-2 patch). I checked that I could boot from it, mount all my filesystems and generally do anything you could do with the standard potato rescue floppy, short of installing from scratch... > You're better off using a separate drive as a root partition for LVM or > soft-RAID. ...Hence, I don't agree with your statement. RAID is for providing resilience, not data protection; tape backups (which, on a server, you should be doing anyway) protect data. If you're running the root fs on a single device and it fails, you'll have to restore the system from backup or rebuild from scratch. After that, restoring the data from backup is a cinch, so you might as well not run RAID at all. Bitter experience speaks, not I. A server we had here, set up with root fs on one disk and data on RAID, failed when the disk with the root fs went kaput. Not only did my predecessor as admin (I was then his assistant) go through ground-up rebuild hell, with all the users on his case, he also had a lot of explaining to do to some very un-impressed bosses, who were asking why they'd paid money for an array of disks when a single disk failure could take the server down for a complete OS rebuild, and why we were using RAID to protect data which was on tape backup anyway, instead of providing resilience. I learnt the lesson well. My servers are now redundant right down to the MBR's; you can use the SCSI BIOS to switch boot disks, and boot from any of the five disks in the array. The machine continues to run, and is able to reboot, after failure of any one disk. I have tested this by simply withdrawing a disk (hot-swap SCSI). You may have to use the BIOS to switch boot disks if the failed disk happens to be the one the machine was configured to boot from, but that's all. All this with potato and kernel 2.2.19. With woody and kernel 2.4.17, I have fond hopes of combining the redundancy achieved with Linux Software RAID with the flexibility of LVM to build a real enterprise-level server. Best regards, George Karaolides
