Hi all,

I'm setting up a machine with a SATA mobile rack in it (StarTech DRW113SAT if anyone's interested), and would like to be able to swap the drives out without rebooting (for backups). My motherboard is an Intel and the SATA chipset is ICH5R. From what I've read, I can't use hotplug with this chipset because it doesn't support it. However, according to http://linux.yyz.us/sata/sata-status.html:

a) "newer Intel ICH" should be able to support hotplug. But more interestingly to me,
b) ""Coldplug" support is potentially feasible." meaning that it should be possible to tell the system you're going to unplug the drive and have it work, rather than just unplugging it. This would be a fine solutionf or me.

If I just unmount and remove the drive, nothing is wrong, but when I put a new drive in and try to mount or fdisk it (even if it's exactly the same as the old one with the exact same partitions) it gives me a bunch of errors. dmesg says:

ATA: abnormal status 0x80 on port 0XEFAF

lots of times, as well as:

Buffer I/O error on device sdb1, logical block x

where x is an integer between 0 and 31.

I am assuming that the first error causes the second one.

So, does anyone have any ideas on getting hotplug or any other option for swapping drives working? For Windows, the drive drawer comes with some software that lets you swap drives, although I haven't tried it since I don't have a Windows machine with SATA so I don't know if it works with this (or any) chipset. A possible workaround would be to reboot the system between swaps (which would be feasible since the swap will happen in the evening and the drive won't be needed again until morning), but I don't want to do that really.

TIA for all suggestions.

Use a hardware SATA card. 3ware or LSI. You are using drivers ala winmodems to assign RAID. There is no hotswapping with software.

Who said anything about RAID? Read before you post.

Bump?

Do a search on the internet about adding and removing SCSI devices. What I seen on the internet, SATA and SCSI is about the same commands when it comes to hotswapping.

Yeah, I did some searching on this yesterday. It would seem that the correct way to remove a SCSI disk is:

echo "scsi remove-single-device 0 0 0 0" > /proc/scsi/scsi

and to add a device is:

echo "scsi add-single-device 0 0 0 0" > /proc/scsi/scsi

(changing 0 0 0 0 to match the location of the drive of course).

I have my removable drives on the second SATA channel, so I did echo "scsi remove-single-device 1 0 0 0" > /proc/scsi/scsi, which worked fine. Now, as long as I don't swap the drive out I can do an echo "scsi add-single-device 1 0 0 0" > /proc/scsi/scsi and it's all peachy. However, if I swap drives before doing the add-single-device, then I get the same errors as in my original post. I even tried removing the little peg from the lock on the rack so that the rack would stay powered on while I was swapping drives - same result.

The ICH5R chipset does claim that it's not capable of hotswapping, however I'm interested in any possible workarounds like the "coldplug" idea mentioned in the link I posted. For now, the server will reboot every night after the backup drives have been swapped but before the next day's backup cycle begins. This works fine in this case, since this is a file/samba server and no one is using it at night, but it's obviously not the preferred option.

You coud get SATA controller like HighPoint RocketRAID 1520 or Highpoint RocketRAID 2220. If you do not want to do that, use USB or Firewire. I suggest Firewire because it has lower latency than USB and maybe better for long write periods.

The store at http://www.cooldrives.com has many USB and Firewire products to choose from.

I'm not looking for RAID support, but if I decide that rebooting the machine every night (and in the very rare instance that I need to do a restore from a differen't day's backup drive) is too cumbersome, I'll pick up a Promise TX4 PCI SATA controller. I'm installing the server this coming Saturday. I haven't touched anything in how I have it configured since I started this thread, and for now it is running just fine with nightly reboots. If this method works as well when the server is in its permanent role (at a different office) as it does right now, then I'll just leave it as it is until I can find some sort of solution (or until the comany it's for outgrows it and needs a new one anyway).

The name of the Highpoint controller just has RAID. It is just a sales attempt just like Promise is saying they are the worldwide leader in SATA RAID. 3ware is better than Promise in the SATA RAID. You do not have to use its RAID although even if you do, you have to use dmraid. Also I have read many problems with Promise controllers. Promise rarely updates their open source drivers with new features that their binary drives have. This is good to know if you thinking of upgrading the kernel in the future.

You are better off using USB or Firewire for backup purposes. I use USB for backups. Backups should done on a different system than the server.

Thanks for the info. I'll look into a controller card if I end up needing one.

As for the backups, I do know what I'm doing. The idea is that there's a SATA drive for each day of the week, each in an HD caddy that can be put in the mobile rack that's in the server. Each night, the person responsible for backups takes out today's drive and takes it off-site, then puts in the drive for the next day. A full backup is done every night, and incrementals every hour during the work day. There's an extra drive in the rotation for a monthly backup. So, at all times we have a backup from the end of the previous month, and a backup from every working hour from the past week. It's an excellent and very fast system (overkill, really, for a company of this size, but that's better than underkill). The only little issue right now is swapping the drives, although rebooting between when the drives are swapped and when the nightly full backup is done works just fine. We also have an identical backup server - if something in the main one dies, the hard drive from the main one can be put into the backup and work can continue as normal while I fix the main server.