Some new findings!
After a little bit of googling I found two issues on github, which are related to my question:
ZFS io error when disks are in idle/standby/spindown mode #4713
0.6.5.6 - I/O timeout during disk spin up #4638
In the first link User antst posted "With 6 drives it takes quite some time to spin-up all drives, because spin-up is sequential."
So I think the staggered disk spinup is a desired behavior of ZFS on Linux.
In both threads two users have written contributions, which are active on this board too.
@luxflow, @hoppel118: I assume that you have posted in the mentioned threads on github. I did a lot of investigations but I found no hint, how to disable the sequential spin-up, if possible at all. Do you have any advice for me? Thanks in advance.
P.S: I could gain no bad experience with disk spindown in ZFS so far. Apart from a 'communication failure' in 'File Systems'. But this is another story in a different thread 
Hello @cabrio_leo
yes, I posted in both of the github threads.
For me it's possible to configure the staggered spin up in the bios of my sas controller (lsi3008).
Which kind of sas/sata controller from which vendor do you use?
Did you check your controller bios?
Greetings Hoppel
Hello @hoppel118
Which kind of sas/sata controller from which vendor do you use?
I am using the onboard SATA connectors, whereof this board has 12 of them. 4 of my 6 disks are connected to the on-board Marvell 88SE9230 storage controller, the rest is connected to the on-board INTEL C2550 SATA controller.
Did you check your controller bios?
Yes, but only the INTEL controller has settings for staggered disk spin-up. It is deaktivated. For the Marvell 9230 I didn´t find any option to change the disk spin-up behavior.
That is why I think staggered disk spin-up should be deaktivated. It would appear to be the case, but it isn´t, particularly if I go to "Storage -> File Systems".
Is it possible for you to connect all your hdds to your intel controller for testing purposes?
Greetings Hoppel
Is it possible for you to connect all your hdds to your intel controller for testing purposes?
Not completely because the OS disk is connected to the intel controller. But for 5 of my 6 data disks this should be possible. But I can make this test on Thursday evening earliest.
What tests should I perform?
I think there is no difference, if you connect 4 of 5 disks to your intel controller. At least for the one disk it is possible to be handled in a different way by the marvel controller.
May be you can try to install all your ZoL disks at your intel controller and check if the deactivated staggered spinup works as expected. It should be possible to connect your system disk to the marvel controller.
If you arrived at this setup, check if your ZoL disks still spin down and spin up staggered. If so I can't help you with deactivating staggered spin up. You deactivated it in the controller bios. I don't think that it is possible for ZoL to control the spin up.
But you can try to disable spin down on all of your disks. Which vendor are your disks from?
In the following post I disabled the spin down of my ZoL disks with idle3tools for western digital disks:
https://github.com/zfsonlinux/…38#issuecomment-252673694
If you don't like that, it's also possible to bring back the default configuration of your disks.
Greetings Hoppel
I disabled the spin down of my ZoL disks with idle3tools for western digital disks:
Hallo @hoppel118,
I am a little bit irritated what you are writing about the WD idle 3 tool. I know this tool in order to change the behavior of the 'IntelliPark'-feature, where the disk heads are parked on the landing zone, when there is no disk access a certain time. On 'green' WD drives the default value is 8 seconds. Under certain conditions this can lead to an enormeous increase of the 'load cycle count', which increase the wear of mechanical parts.
For all of my WD Red drives I have disabled it completely.
But the spin down of a drive is normaly not affected by the idle3-tool - as far as I know.
At github I described what I did with my wd red disks. I wrote that I deactivated the default spindown of the disks with idle3tools and it worked.
But maybe I understood something wrong. It's possible. 
Greetings Hoppel
I don't think that it is possible for ZoL to control the spin up.
@hoppel118 I think you are **sooooo** right! And I was so mistaken! There is no staggered disk spin up in my setup!
I did some further investigations. It took some time to detect that this staggered disk spin up only happens when I go to 'Storage -> File Systems' in the WebUI. Then the disks start one after another from standby mode.
When the disks are in standby and I press <F5> in order to refresh a Samba Share in Windows for example, all disks start simulanously and the share is reachable again after a short time. This is exactly what I want.
So there is really no problem. Therefore I will set the thread state to solved.
Great, that you have a proper solution. Did you change something else?
Did you change something else?
No, now I am back on the default configuration.
Before I was just about to connect all data disks to the INTEL controller, as you have recommended. But the Node 304 case is quite small and there is only little space to move. It was difficult to accomodate all cables then, because the connectors are partially covered. I would prefer not to change my hardware setup frankly.
But I did a short test with modified spin up settings for the two disks connected to the INTEL controller. But there was no difference. The same behavior as I have discribed it in my post no. 10 above.
Sie haben noch kein Benutzerkonto auf unserer Seite? Registrieren Sie sich kostenlos und nehmen Sie an unserer Community teil!
