[smartmontools-support] Long test heavy on I/O

Carlos E. R. robin.listas at telefonica.net
Tue Nov 7 15:54:32 CET 2017


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



On Saturday, 2017-11-04 at 10:38 +0100, Gandalf Corvotempesta wrote:

> Hi to all
> I've seen, on multiple servers, with multiple disks (all SAS, all
> Seagate 15k) that running a smart long test is resulting in very high
> I/O and will slow down any other disk access.
> 
> Based on what I've read on Net, smart long test should be made
> automatically with low priority by disk itself, but is not our case.
> 
> Every time we run a long test on a disk (member of an hardware RAID),
> any VM reading/writing to that disk will show load arout 80-90 and
> after a while, some crashes.
> 
> "iostat -x" show 100% util for that disk.
> 
> If I stop the long test (smartctl -X) load immediatly goes down to 1-2
> (as it should be) and %util from "iostat" will decrease to about
> 30-40%
> 
> Re-running the long test will make the same.
> 
> Other than this, a long test is expected to finish in about 2 hours
> (based on what smart is saying) but after 8 hours is still running.

My guess is that the long test impacts badly disk I/O from other tasks 
that run on the computer, as the test itself does not use I/O at all, it 
runs inside the disk. The computer has to wait on the disk to become 
available.

Of course, as the test is interrupted (low priority) when the computer 
demands access, it takes many hours to complete.


- -- 
Cheers,
        Carlos E. R.
        (from openSUSE 42.2 x86_64 "Malachite" at Telcontar)

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2

iEYEARECAAYFAloBySgACgkQtTMYHG2NR9X8+gCeJ4zIHfwRbRoYaTK7lDTtpFtu
leMAmgKny4Qww7eaeTSrKxF7GeAd/b8o
=as7X
-----END PGP SIGNATURE-----



More information about the Smartmontools-support mailing list