ZFS RAIDZ and atacontrol spindown settings

[Mark Stapper]

Arnaud Houdelette wrote:
> Mark Stapper wrote:
>> Arnaud Houdelette wrote:
>>  
>>> Mark Stapper a écrit :
>>>    
>>>> Ivan Voras wrote:
>>>>  
>>>>      
>>>>> 2009/7/21 Mark Stapper <stark at mapper.nl>:
>>>>>             
>>>>>> Ivan Voras wrote:
>>>>>>                   
>>>>>>> Mark Stapper wrote:
>>>>>>>                         
>>>>>>>> Good day,
>>>>>>>>
>>>>>>>> I am the proud user of a FreeBSD 7.2 AMD64 system housing,
>>>>>>>> amongst other
>>>>>>>> things, a data server.
>>>>>>>> My "server"(It's called "Yoshi") runs FreeBSD from a mirrored
>>>>>>>> system
>>>>>>>> disc, and has a zfs RAIDZ array with 4 discs for bulky data.
>>>>>>>> As it is a home server, and I work during the day, these four
>>>>>>>> discs were
>>>>>>>> spinning happily all day long without much use for them doing so.
>>>>>>>> To save the world(and money) I issued the command "atacontrol
>>>>>>>> spindown
>>>>>>>> 1800" for all the discs in my array spinning them down after
>>>>>>>> thirty
>>>>>>>> minutes of idle spinning.
>>>>>>>> So far so good, me very happy! However, when I access the array
>>>>>>>> after
>>>>>>>> the discs have been spun down, it spins up the discs one after the
>>>>>>>> other...
>>>>>>>> Mind you one AFTER the other, taking 4*9 seconds to do a "ls"
>>>>>>>> command on
>>>>>>>> my music directory.
>>>>>>>> Content as I am with the smooth down- and upspinning of the
>>>>>>>> disks, I
>>>>>>>> would like it better if the four discs would spin up
>>>>>>>> simultaneously.
>>>>>>>> Thus my  question: "Is it possible to \"group\" discs to be
>>>>>>>> spun up
>>>>>>>> together, or to issue a custom command upon upspinning of a disc
>>>>>>>> such as
>>>>>>>> to spin up other disks?"
>>>>>>>>                                 
>>>>>>> Good question but the answer is probably no - it really only
>>>>>>> depends
>>>>>>> on how ZFS accesses the drives; if it accesses them in sequence,
>>>>>>> you
>>>>>>> can't change it.
>>>>>>>                           
>>>>>> I've been looking at writing a shell script which monitors
>>>>>> /var/log/messages.
>>>>>> something like:
>>>>>>
>>>>>> If last line in /var/log/messages is like "request while spun down.
>>>>>> Starting."
>>>>>>    spinup disks
>>>>>>
>>>>>> couple of problems though, I should probably poll the kernel
>>>>>> messages
>>>>>> every second or so, but if I only check the last linee, I could
>>>>>> miss the
>>>>>> spinup message.
>>>>>> I could count the number of lines in /var/log/messages and keep
>>>>>> count of
>>>>>> the number of lines i've seen. Problem with this approach is that
>>>>>> it's
>>>>>> not very efficient.
>>>>>> So I was hoping there is a way to receive this kernel message
>>>>>> directly.
>>>>>> I am now thinking in the lines of a program which received a
>>>>>> signal on
>>>>>> new kernel messages available or something similar.
>>>>>>
>>>>>> Any thoughts?
>>>>>>                     
>>>>> You could do what "tail -f" does and simply hook a kqueue to get new
>>>>> messages from the /var/log/messages file.
>>>>>               
>>>> I'll try that! thanks!
>>>>
>>>>         
>>> As I get quite the same issue with a 4 disk raidz pool, could you
>>> please tell me the result of your tries ? which command do you use to
>>> force the disks to spin up ?
>>>
>>> Thanks
>>>     
>> I have been using a shell script containing:
>> #!/bin.sh
>> dd if=/dev/ad4 of=/dev/null count=1&
>> dd if=/dev/ad6 of=/dev/null count=1&
>> dd if=/dev/ad8 of=/dev/null count=1&
>> dd if=/dev/ad10 of=/dev/null count=1&
>>   
>
> Thanks.
>
> FYI, I made the little script behind to speed up the spinup of the
> whole pool, in case it be some use for anybody :
>
> #!/bin/sh
> wait-spinup() {
> while true
> do
> read l
> l=`echo $l | grep 'request while spun down'`
>  if [ -n "$l" -a "`date +%H%M%S`" != "$d" ]; then
>    d=`date +%H%M%S`
>    dd if=/dev/ad4 of=/dev/null count=1 2> /dev/null &
>    dd if=/dev/ad6 of=/dev/null count=1 2> /dev/null &
>    dd if=/dev/ad8 of=/dev/null count=1 2> /dev/null &
>    dd if=/dev/ad10 of=/dev/null count=1 2> /dev/null &
>  fi
> done
> }
>
> tail -n 0 -F /var/log/messages | wait-spinup
>
>
Very very nice :-)
Before:
yoshi# time sh ./spinup.sh
 0.000u 0.007s 0:39.13 0.0%     0+0k 0+0io 0pf+0w

/var/log/messages
Jul 23 09:05:56 yoshi kernel: ad4: request while spun down, starting.
Jul 23 09:06:06 yoshi kernel: ad6: request while spun down, starting.
Jul 23 09:06:16 yoshi kernel: ad8: request while spun down, starting.
Jul 23 09:06:25 yoshi kernel: ad10: request while spun down, starting.

With your script:

yoshi# time sh ./spinup.sh
0.000u 0.007s 0:10.10 0.0%      0+0k 0+0io 0pf+0w

/var/log/messages
Jul 23 09:04:38 yoshi kernel: ad4: request while spun down, starting.
Jul 23 09:04:38 yoshi kernel: ad6: request while spun down, starting.
Jul 23 09:04:38 yoshi kernel: ad8: request while spun down, starting.
Jul 23 09:04:38 yoshi kernel: ad10: request while spun down, starting.

Finally i won't have to wait 40 seconds for a directory listing anymore!
Thank you very very much! :-)
Greetz,
Mark



-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 259 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20090723/70d63fdf/signature.pgp