AHCI timeout when using ZFS + AIO + NCQ
Vladislav Prodan
universite at ukr.net
Sun Jan 27 15:29:05 UTC 2013
> ----- Original Message -----
> From: "Vladislav Prodan" <universite at ukr.net>
>
> >> Is it always the same disk, of so replace it SMART helps identify issues
> >> but doesn't tell you 100% there's no problem.
> >
> >
> > Now it has fallen off a different HDD - ada0.
> > I'm 99% sure that MHDD will not find problems in HDD - ada0 and ada2.
> > I still have three servers with similar chipsets that have similar problems
> > with blade ahci times out.
>
> I notice your disks are connecting at SATA 3.x, which rings bells. We had
> a very similar issue on a new Supermicro machine here and after much
> testing we proved to our satisfaction that the problem was the HW.
I have a motherboard ASUS M5A97 PRO
http://www.asus.com/Motherboard/M5A97_PRO/#specifications
Has replacement SATA data cables.
Putting hard RAID controller does not guarantee data recovery at his death.
> Essentially the combination of SATA 3 speeds the midplane / backplane
> degraded the connection between the MB and HDD enough to cause
> the disks to randomly drop when under load.
>
> If we connected the disks directly to the MB with SATA cables the
> problem went away. In the end we had midplanes changed from an
> AHCI pass-through to active LSI controller.
>
> So if you have any sort of midplane / backplane connecting your disks
> try connecting them direct to the MB / controller via known SATA 3.x
> compliant cables and see if that stops the drops.
>
> Another test you can do is to force the disks to connect at SATA 2.x
> this also fixed it in our case, but wasn't something we wanted to
> put into production hence the controller swap.
>
> To force SATA 2 speeds you can use the following in /boot/loader.conf
> where 'X' is disk identifier e.g. for ada0 X = 0:-
> hint.ahcich.X.sata_rev=2
>
> Hope this helps.
>
> Regards
> Steve
>
--
Vladislav V. Prodan
System & Network Administrator
http://support.od.ua
+380 67 4584408, +380 99 4060508
VVP88-RIPE
More information about the freebsd-current
mailing list