* Change the weight of osd
@ 2011-05-06 6:39 doki74216
2011-05-09 2:08 ` doki74216
0 siblings, 1 reply; 3+ messages in thread
From: doki74216 @ 2011-05-06 6:39 UTC (permalink / raw)
To: Ceph
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=Big5, Size: 3069 bytes --]
Dear Ceph,
I am trying changing the weight of OSDs.
My testing environment is one MON , one MDS and two OSDs(named OSD9 and OSD10).
There are 4 disk foe each server host.
In each OSD host, I running one cosd process per disk.
Therefore, there are 4 cosd processes in OSD9 and OSD10 separetly.
I named the cosd processes as osd24 osd25 and osd26 in OSD9.
I also named the cosd processes as osd27 osd28 and osd29 in OSD10.
My testing steps as shown below, if there is any wrong please correct
me as far as possible. Thank you.
1) I want to compare the size of the folder whose weight is changed
before and after.
So I record the size of the folder before the weight of osd is changed .
I excute"du /mnt/ext4/osd24/current/"
...
...
"du /mnt/ext4/osd29/current/"
2) I edit the /tmp/crushmap.txt. Here is the txt.
====================================
domain dom8 {
id -9 # do not change unnecessarily
alg uniform # do not change bucket size (3) unnecessarily
hash 0 # rjenkins1
item device24 weight 1.000 pos 0
item device25 weight 1.000 pos 1
item device26 weight 1.000 pos 2
}
domain dom9 {
id -10 # do not change unnecessarily
alg uniform # do not change bucket size (3) unnecessarily
hash 0 # rjenkins1
item device27 weight 1.000 pos 0
item device28 weight 1.000 pos 1
item device29 weight 3.000 pos 2
^^^^^^^^
I replace 1.000 with 3.000 .
======================================
3) I execute "crushtool -c /tmp/crushmap.txt -o /tmp/crushmap_new¡¨and¡¨
ceph osd setcrushmap -i /tmp/crushmap_new¡¨
4) I thought the crushmap has already changed.
If I write data, the data will more prbable placed in osd29 . right?
When I excute" time dd if=/dev/zero
of=/mnt/ceph/reweight_64k_64M_1.dbf bs=64k count=1024 oflag=dsync"
I check the folder size by excuting "/mnt/ext4/osd24/current/"....
"/mnt/ext4/osd29/current/".
I found that the folder size is not growed in accordance with the
wight I edit (osd24 osd25 osd26 osd27 osd28 weight=>1.000, osd29
weight=>3.000).
Here is the increasing folder size.
In OSD9
Ć osd24 4124 kB
Ć osd25 28748 kB
Ć osd26 34064 kB
In OSD10
Ć osd27 4124 kB
Ć osd28 28748 kB
Ć osd29 34060 kB
^^^^^^^^^^^^^^^^^^
Why the folder size is not growed in accordance with
the wight I edit ?
If I want to change the weight of osd . Are the steps I mentioned above correct?
Do you have any good idea or advice to our test? I really want to know
how to change the weight of osd and how to verify it.
Thank you very much...^^
Stefanie Chen
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Change the weight of osd
2011-05-06 6:39 Change the weight of osd doki74216
@ 2011-05-09 2:08 ` doki74216
2011-05-09 23:42 ` Sage Weil
0 siblings, 1 reply; 3+ messages in thread
From: doki74216 @ 2011-05-09 2:08 UTC (permalink / raw)
To: Ceph
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=Big5, Size: 3691 bytes --]
Dear developer,
There is a slip of the pen in last letter.
To avoid you misunderstand what I describe.
Here is my correction..
¦b 2011¦~5¤ë6¤é¤U¤È2:39¡Adoki74216@gmail.com <doki74216@gmail.com> ¼g¹D¡G
> Dear Ceph,
> I am trying changing the weight of OSDs.
> My testing environment is one MON , one MDS and two OSDs(named OSD9 and OSD10).
> There are 4 disk foe each server host.
^^^^ not foe, I correct is as "for"
> In each OSD host, I running one cosd process per disk.
> Therefore, there are 4 cosd processes in OSD9 and OSD10 separetly.
^^^^^ not 4 , I correct it as "3". There
are one disk for OS system and 3 disk for osd.
> I named the cosd processes as osd24 osd25 and osd26 in OSD9.
> I also named the cosd processes as osd27 osd28 and osd29 in OSD10.
>
> My testing steps as shown below, if there is any wrong please correct
> me as far as possible. Thank you.
>
> 1) I want to compare the size of the folder whose weight is changed
> before and after.
> So I record the size of the folder before the weight of osd is changed .
> I excute"du /mnt/ext4/osd24/current/"
> ...
> ...
> "du /mnt/ext4/osd29/current/"
> 2) I edit the /tmp/crushmap.txt. Here is the txt.
> ====================================
> domain dom8 {
> id -9 # do not change unnecessarily
> alg uniform # do not change bucket size (3) unnecessarily
> hash 0 # rjenkins1
> item device24 weight 1.000 pos 0
> item device25 weight 1.000 pos 1
> item device26 weight 1.000 pos 2
> }
> domain dom9 {
> id -10 # do not change unnecessarily
> alg uniform # do not change bucket size (3) unnecessarily
> hash 0 # rjenkins1
> item device27 weight 1.000 pos 0
> item device28 weight 1.000 pos 1
> item device29 weight 3.000 pos 2
> ^^^^^^^^
> I replace 1.000 with 3.000 .
> ======================================
> 3) I execute "crushtool -c /tmp/crushmap.txt -o /tmp/crushmap_new"and"
> ceph osd setcrushmap -i /tmp/crushmap_new"
> 4) I thought the crushmap has already changed.
> If I write data, the data will more prbable placed in osd29 . right?
> When I excute" time dd if=/dev/zero
> of=/mnt/ceph/reweight_64k_64M_1.dbf bs=64k count=1024 oflag=dsync"
> I check the folder size by excuting "/mnt/ext4/osd24/current/"....
> "/mnt/ext4/osd29/current/".
> I found that the folder size is not growed in accordance with the
> wight I edit (osd24 osd25 osd26 osd27 osd28 weight=>1.000, osd29
> weight=>3.000).
> Here is the increasing folder size.
> In OSD9
> Ć osd24 4124 kB
> Ć osd25 28748 kB
> Ć osd26 34064 kB
> In OSD10
> Ć osd27 4124 kB
> Ć osd28 28748 kB
> Ć osd29 34060 kB
> ^^^^^^^^^^^^^^^^^^
> Why the folder size is not growed in accordance with
> the wight I edit ?
>
>
> If I want to change the weight of osd . Are the steps I mentioned above correct?
> Do you have any good idea or advice to our test? I really want to know
> how to change the weight of osd and how to verify it.
> Thank you very much...^^
>
> Stefanie Chen
>
Please helping me to solve this problem.
I appreciate it.
Stefanie Chen
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Change the weight of osd
2011-05-09 2:08 ` doki74216
@ 2011-05-09 23:42 ` Sage Weil
0 siblings, 0 replies; 3+ messages in thread
From: Sage Weil @ 2011-05-09 23:42 UTC (permalink / raw)
To: doki74216; +Cc: Ceph
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=Big5, Size: 2932 bytes --]
Hi Stefanie,
On Mon, 9 May 2011, doki74216@gmail.com wrote:
> > domain dom8 {
> > id -9 # do not change unnecessarily
> > alg uniform # do not change bucket size (3) unnecessarily
^^^^^^^
The problem is that the bucket algorithm is 'uniform', which uniformly
distributes objects across the items. Weighting isn't supported, and the
weight value is ignored. (We should make the crushmap compiler spit out a
warning to that effect if the weights vary at all!)
If you switch the alg to straw, the data will reshuffle, but you'll be
able to set the weights however you like.
sage
> > hash 0 # rjenkins1
> > item device24 weight 1.000 pos 0
> > item device25 weight 1.000 pos 1
> > item device26 weight 1.000 pos 2
> > }
> > domain dom9 {
> > id -10 # do not change unnecessarily
> > alg uniform # do not change bucket size (3) unnecessarily
> > hash 0 # rjenkins1
> > item device27 weight 1.000 pos 0
> > item device28 weight 1.000 pos 1
> > item device29 weight 3.000 pos 2
> > ^^^^^^^^
> > I replace 1.000 with 3.000 .
> > ======================================
> > 3) I execute "crushtool -c /tmp/crushmap.txt -o /tmp/crushmap_new"and"
> > ceph osd setcrushmap -i /tmp/crushmap_new"
> > 4) I thought the crushmap has already changed.
> > If I write data, the data will more prbable placed in osd29 . right?
> > When I excute" time dd if=/dev/zero
> > of=/mnt/ceph/reweight_64k_64M_1.dbf bs=64k count=1024 oflag=dsync"
> > I check the folder size by excuting "/mnt/ext4/osd24/current/"....
> > "/mnt/ext4/osd29/current/".
> > I found that the folder size is not growed in accordance with the
> > wight I edit (osd24 osd25 osd26 osd27 osd28 weight=>1.000, osd29
> > weight=>3.000).
> > Here is the increasing folder size.
> > In OSD9
> > Ć osd24 4124 kB
> > Ć osd25 28748 kB
> > Ć osd26 34064 kB
> > In OSD10
> > Ć osd27 4124 kB
> > Ć osd28 28748 kB
> > Ć osd29 34060 kB
> > ^^^^^^^^^^^^^^^^^^
> > Why the folder size is not growed in accordance with
> > the wight I edit ?
> >
> >
> > If I want to change the weight of osd . Are the steps I mentioned above correct?
> > Do you have any good idea or advice to our test? I really want to know
> > how to change the weight of osd and how to verify it.
> > Thank you very much...^^
> >
> > Stefanie Chen
> >
>
>
> Please helping me to solve this problem.
> I appreciate it.
> Stefanie Chen
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2011-05-09 23:41 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-05-06 6:39 Change the weight of osd doki74216
2011-05-09 2:08 ` doki74216
2011-05-09 23:42 ` Sage Weil
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.