Re: [Xen-users] Cheap IOMMU hardware and ECC support importance

[Gordan Bobic <gordan@xxxxxxxxxx>]

On 07/10/2014 03:34 AM, lee wrote:
> Nuno MagalhÃes <nunomagalhaes@xxxxxxxxx> writes:
>
> > On Wed, Jul 9, 2014 at 3:29 AM, lee <lee@xxxxxxxxxxxxxxx> wrote:
> > > It's simple with RAID because the disk either fails or not.  It's
> > > usually simple with disks because it either fails or not.
> > Depends on your RAID, depends on your disks. If you use disks without
> > TLER they may get kicked out of the array even if they're not failing.
> That brings back the question of when a disk has actually failed.
It also highlights that disk "failure" is often nowhere nearly as clear 
cut as most people realize.
> Should I continue to entrust my data to a disk that spends long times on
> trying to correct errors?
Any disk without TLER (Time Limited Error Recovery) and having it 
enabled will sometimes do that. You also need a RAID controller that 
either enables this feature on the disk itself or exposes the raw disk 
in a way that you can do it yourself, as the setting defaults to off. It 
wouldn't surprise me if various storage vendors ship disks with doctored 
firmwares that, among other things, have this setting defaulting to off. 
Some disk enclosures even go as far as detecting the disks that are 
plugged in, and if it recognizes the model number it'll flash the disk's 
firmware to the certified version (the enclosure has a built in library 
of certified firmwares for certified disk models). I only found out 
about this feature of some enclosures when a client got some disks that 
seemed legit but turned out to be "fake" (in the sense that they were 
desktop drives doctored to say they were enterprise drives). So when the 
enclosure flashed the certified enterprise firmware onto them they got 
bricked because the drive electronics were actually different).
> Should I use it as a spare for a RAIDZ-1
> volume and hope that it might be helpful?
With ZFS you might be able to get away with it by bumping the SCSI 
command timeout to however long the disk takes to time out the 
operation. That will prevent the disk getting kicked out of the pool, 
but you'll have to live with all pool operations blocking until that 
disk responds.
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users