Hi, Richard!
Just curious, what would "iostat 1 10" and "vmstat 1 10" return ?
Best regards,
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] De la part de Richard Stockton
Envoyé : jeudi 19 mai 2011 00:08
À : Felix Frank
Cc
Hi Jan !
This is a tough question. It depends on many parameters, starting with the
size of the requested IOs. There are a couple of good articles on the net
talking about that, including these :
- 2 from Scott LOWE, wich are fairly clear and interesting from the
IOPS estimation metho
Hi, Richard!
Ok, good news for the NICs, but bad news to you unfortunately :)
Well, since the /dev/sdb is used by this /work directory, we might try to
investigate this track. Some infos missing though : When the platform calms
down, do you still see à 10:1 ratio between sdb and other volumes writ
Hello Chris!
Ok, these numbers make more sense, your platform is sane. However this shows
something very interesting about what's happening under the cover : The IO
profile is different on both side, the destination issuing smaller IOs. In
order to get something optimized, I'd rather expect the IO
mment ?
Best regards,
Pascal.
-Message d'origine-
De : Gouveia, Chris [mailto:chris.gouv...@spirent.com]
Envoyé : lundi 20 juin 2011 20:54
À : Pascal BERTON; drbd-user@lists.linbit.com
Objet : RE: [DRBD-user] Secondary server works harder than primary
Does anyone have any ideas wha
Hi Richard!
Glad to read that! Wow, I would never had thought about that, so I write it
somewhere in my head, very good to know! And it makes so much sense by the
way... Thanks for sharing this.
Best regards,
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:dr
And mostly don't forget to set $LANG to "C" ! ;)
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] De la part de Lars Ellenberg
Envoyé : jeudi 28 juillet 2011 17:48
À : drbd-user@lists.linbit.com
Objet : Re: [DRBD-user] Unterschiedlich
Pascal,
One thing is unclear : did it used to work in the past (and if yes what has
changed lately that could explain this behavior) or is it a new feature you’ve
just added to your customer’s config ?
Furthermore, I suspect you have scripted all this process haven’t you ? If so,
have you i
Hi Christian!
To what I understand in your mail, the key fact here is that your VM
(Primary, correct ?) has booted with an updated kernel. You don't explain
why, was it a human, wanted operation ? Nevertheless I just can't figure out
any other one in fact...
Well, so the kernel has changed... So t
Nick,
The 2 URLs (Which in fact are just one same article) are interesting.
However, the setup that this guy describes looks fairly simple, and mostly
"home lab" oriented to what I understand. In my personal case, that's
something to put in production that I need, and production needs something
pr
Hi Dan!
VMFS is a clustered filesystem by nature, in the sense it allows concurrent
read-write accesses from up to 32 nodes simultaneously, it's not
specifically as of version 5.
Unfortunately you (still) have no mean to replicate it safely in
dual-primary mode with DRBD like you would do with OCF
Hi, Kankamuso!
Following up Digimer's reply, at HP the Proliant DL180g6 is also a good
model I think : It's available in 25 hot-plug 2,5" HDD slots (or 4, 8 or 12
3.5" slots, but I personnaly prefer the 25 slot version). It's a 2 way Intel
server, able to host up to 192GB RAM. One of its best feat
Hi Elias !
crm status will tell you on which node a given resource is active. You can
also use crm_mon (underscore!) which will present the same thing in real
time (crm status is a one shot run).
Basically, crm is the command to use to do everything you intend to do.
Regarding the iSCSI t
Oren,
AFAIK, DRBD doesnt support snapshotting natively. Therefore, if Im
correct, you only have 2 ways to achieve what you intend to do :
- Setup an VG/LV structure on top of your DRBD resource and make
use of the LVM snapshot facility. Beware, Ive read many articles mentioning
tha
Florian,
I've watched your video, very interesting indeed... Unfortunately for you it
raised some more questions to me :-)
First, just to be sure I correctly understood : flashcache will only (But
significantly) improve read operations, no impact or so on writes right ?
Then, a year or 2 ago I had
>On Sat, Feb 11, 2012 at 7:30 PM, Pascal BERTON
wrote:
>> Florian,
>>
>> I've watched your video, very interesting indeed... Unfortunately for you
it
>> raised some more questions to me :-)
>> First, just to be sure I correctly understood : flashcache will on
blem, but the incorrect statement needed
illumination.
Dan in Atlanta
-Original Message-
From: drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] On Behalf Of Pascal BERTON
(EURIALYS)
Sent: Friday, February 17, 2012 6:52 AM
To: 'Mark Watts'; drbd-user@li
Guys,
I'm trying to better understand the unplug-watermark parameter settings, and
honestly I don't find much info about a rational way to do it. I understand
that it strongly relies on the general specs of the storage subsystem in
use, but except that. I've seen a discussion about the CCISS dr
Wow! Sandy Bridge for this ? It's like taking a shotgun to kill a flee,
isn't it ? Wouldn't you do better placing your money on your disk layout,
rather than on the CPU family ? That's what I'd do at least!
What do you intend to do with this platform then ? Take care : Few disks,
few IOs, especiall
Dirk,
Isn't there an error in your numbers ?
You say that with you non-custered system you transfer 500GB in 400
seconds, which means more than a GB per second, and I can't figure out
how you can achieve that with 100Mb link that can't run faster than
roughly 10MB/s. Or did I miss something ?
Ah Ok!
Le 04/04/2012 14:19, Kaloyan Kovachev a écrit :
On Wed, 04 Apr 2012 14:12:02 +0200, Pascal BERTON
wrote:
Dirk,
Isn't there an error in your numbers ?
You say that with you non-custered system you transfer 500GB in 400
seconds, which means more than a GB per second, and I can
Thanks for the tip guys, I'm used to unison, not rsync...
So, no idea, then... :-)
Le 04/04/2012 14:20, Tristan de CACQUERAY a écrit :
On Wed, 04 Apr 2012 14:12:02 +0200
Pascal BERTON wrote:
Dirk,
Isn't there an error in your numbers ?
You say that with you non-custered system yo
Dan,
Isn't it the backing device you should refer ? To me, if you want multipath
to work correctly, the device has to return an ID. Having said that, if I
try it on my resource having the following config file :
resource vmfs2 {
device /dev/drbd1;
disk /dev/vg_vmfs2/lv_vmfs2;
Stefan,
If the master fails, then the slave is promoted master and takes the
service over. At this time, it's the only node alive. When the former
master returns to life, its data will be outdated compared to the
current master. When it joins again the cluster, DRBD will detect that
inconsist
Hi Matthew!
I've recently experienced the very same behavior, with two bonded 10GbE
direct links between nodes for replication. The nodes host 4 resources
under DRBD 8.3.11 using protocol B, and just like you, disconnections are
intermittent and on any resources, no logical rule for victim electi
Keith,
I'm not sure it's a good idea, however the first thing to avoid is to
present the replicated disk space to your virtualized peers as vDisks on a
VMFS. Use RDMs instead, in physical mode if possible, to grab some more
perfs.
Then for this RDM, use a vdisk node that will cause the instantiati
Guys,
It seems that the drbd.conf man page section (DRBD 8.3 Users Guide page 135)
provides an erroneous max value for the "al-extents" parameter. It reads
3843 when it seems to be 3833 in reality :
[root@euriastore2 drbd.d]# drbdadm create-md vmfs
drbd.d/global_common.conf:27: al-exte
I also used this manual to startup, I agree with Jake, that's a typo. Apart
from that, William, your "bindnetaddr" parameter should not end with a "0",
it's supposed to be the IP address your local host/node will use to monitor its
peers. Instead, replace it by the IP address your server has on
udi 7 juin 2012 09:36
À : Pascal BERTON
Cc : drbd-user@lists.linbit.com
Objet : Re: [DRBD-user] DRBD 8.3 User's Guide - al-extents typo
Hi,
On 06/06/2012 10:24 PM, Pascal BERTON wrote:
> Well, not a big deal, it effectively works fine with 3833
>
> BTW, Ive often seen references to
-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] De la part de Arnold Krille
Envoyé : jeudi 7 juin 2012 21:02
À : drbd-user@lists.linbit.com
Objet : Re: [DRBD-user] Corosync Configuration
On 07.06.2012 19:28, Pascal BERTON wrote:
> I also used this manual to startup,
Hi Keith!
It does. Don't forget to shrink your filesystem(s) first. You'll find a
whole lot of howtos on the net to achieve that depending on the filesystem
you actually use.
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] D
/dev/sda2 ? Wouldn't it sound like a swap partition or something alike ?
Or wouldn't it be already in use by one of the MDs ?
What does "fdisk -l /dev/sda" return ?
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] De la part de Digimer
Hi Simon.
AFAIK, the Ping Ack error means your replication network links are either
down or subject to sufficient errors to prevent both nodes to reach each
other in a timely manner. I had the occasion to experience such behavior
because of bad optical fibers for instance, generating huge numbe
: 'Pascal BERTON'; drbd-user@lists.linbit.com
Objet : RE: [DRBD-user] Drbd : PingAsk timeout, about 10 mins.
Hi Pasical,
Thanks your reply.
Yes, the network was bad. Master host was dead so that Slave host took over
its work and mount the drbd partition on Slave host. When moun
Simon,
You have only sent me the results of "drbdadm 0 show", you forgot to send me
the result of "crm configure show" that I had also asked, could you please send
it too ? I also would like to see what "ifconfig" of your 2 replication
interfaces do report (The ones on subnet 172.17). Please se
Yes, Simon, thanks for the yesterday's ifconfig reports (Which did not
reveal anything wrong BTW), but I also asked you the "crm configure show"
results and you still omitted it...
Regards,
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@list
Felix,
This one has been of great help to me facing very odd errors over 10Gb
replication links. It's been 80% of the resolution of my problems.
http://www.kernel.org/doc/ols/2009/ols2009-pages-169-184.pdf
Regards,
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.com
[ma
Guys,
I'm currently updating a Pacemaker+drbd+scst cluster from drbd-8.3.12 to
drbd-8.3.15, and I found something odd regarding unfencing. Basic symptom is
that unfencing does nothing, cluster logs report unfencing had been done,
but in reality nothing has occurred.
For instance here's the
same version on both nodes, the
rest of the cluster stack being identical.
If someone has any advice or idea, hes welcome!
Regards,
Pascal.
De : drbd-user-boun...@lists.linbit.com
[mailto:drbd-user-boun...@lists.linbit.com] De la part de Pascal BERTON
Envoyé : dimanche 24 février
Objet : Re: [DRBD-user] drbd 8.3.15 - crm_fence_peer.sh - problem in
fence_peer_init()
On Mon, Feb 25, 2013 at 01:09:16AM +0100, Pascal BERTON wrote:
> Well, thinking about it, the only conclusion that I can honestly make
> after this instrumentation is that $id_prefix is effectively incorr
Are you using iSCSI to access your volumes ? Might worth it activating iSCSI
digests on both sides and see how it behaves then, wouldn’t it ? You’d probably
lose some perfs but it would probably too help you identify the root cause of
your problems I guess…
Regards,
Pascal.
De : drbd-us
Piotr,
Reading your infos, I understand that /dev/md2 is at the same time the backing
device of vg1 (pvdisplay recognizes md2 as a valid PV) and drbd1, which is
nonsense.
I'd rather expect you put vg1 on top of drbd1, not md2, or you create 2
resources drbd1 and drbd2 that would use vg1-vm1 and
Hi Antonio !
As far as I know, virtio is the IO layer of KVM. In your case you’re planning
to use ESXi, so I believe you may disregard that point.
However, as Cesar said, performance will be key. So take care to implement the
most performant paravirtualized drivers within your VMs, namely vmx
Maurits,
I faced this issue some times ago. After a few hours scratching my head, my
eyes stopped by luck on one small line in /var/log/messages among the whole
rest, warning me that my 10Gb Ethernet card was requiring a 8x PCIe slot
when it was in a 4x PCIe slot at that moment, and that might hav
[mailto:fred.demarcy...@gmail.com]
Envoyé : mercredi 1 février 2012 17:20
À : Pascal BERTON (EURIALYS)
Cc : drbd-user@lists.linbit.com
Objet : Re: [DRBD-user] Slower disk throughput on DRBD partition
Hi Pascal
1) Both vdisks for /dev/sda and /dev/sdb are on the same datastore which is
made of
Mark,
Snapshot creation requires that VMware tools are installed, which brings the
sync driver onboard. "Cannot quiesce VM" means it is unable to contact the
sync driver. Have you installed the VMware Tools ?
Best regards,
Pascal.
-Message d'origine-
De : drbd-user-boun...@lists.linbit.
46 matches
Mail list logo