[opensuse] Software RAID..
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know.. Anders. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb >in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
software RAID only uses idle bandwidth, so if you are doing stuff, the sync will suffer, also, since they are PATA disks, are they on different channels? (Will make a big difference) Phil -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
At 10:58 PM 2/9/2009 +0100, Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
That's the main reason we abandoned software Raid many years ago - seems like it hasn't improved! The main problem we saw was that in the event of an unplanned power interruption, we saw almost a TWO HOUR parity rebuild in an 80GB array. Rsync is much more friendly. Lee -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
L. V. Lammert schrieb:
At 10:58 PM 2/9/2009 +0100, Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
That's the main reason we abandoned software Raid many years ago - seems like it hasn't improved! The main problem we saw was that in the event of an unplanned power interruption, we saw almost a TWO HOUR parity rebuild in an 80GB array. Meanwhile mdraid keeps track about what's in sync. So such a resync after a power interruption will be much faster.
Only the initial sync is slow. And that's intended, because it shall not hook system resources. Think, the initial sync may even happen during installation!!! But even that's not so bad as it looks, because you have to wait this time only until you reach you planned level of security, but while it's syncing you can use the machine already. You can even reboot or power it down during this phase and resync will start after the next boot and continue, where it has been interrupted. The initial sync will last longer in real time this way, but it will work. Two hours for 80GB is possible for the initial sync, but 40 Hours for 200GB is definitely too much.
Rsync is much more friendly. ... but it's a completely different tool for a different purpose. Don't compare pears with apples. And even for rsync it may be useful to reduce it's bandwitdh.
Herbert -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
L. V. Lammert wrote:
That's the main reason we abandoned software Raid many years ago - seems like it hasn't improved!
As usual YMMV. Anders' system is apparently under such heavy load that it is slowing down the resync a lot. Add to that that he's using a fairly slow disk/interface ...
Rsync is much more friendly.
So is my cat, but what's that got to do with it? /Per -- Per Jessen, Zürich (7.1°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Tue, 10 Feb 2009, Per Jessen wrote:
L. V. Lammert wrote:
Rsync is much more friendly.
So is my cat, but what's that got to do with it?
Not all cats are friendly, .. but rsync always does a nice job. No overhead (unless you're doing a sync), on failure the system can be immediately brought back up on the other drive, it can also be used to manage versions a la rsnapshot. Sounds a lot more friendly [& useful] than some cats. Lee ============================================== Leland V. Lammert lvl@omnitec.net Chief Scientist Omnitec Corporation Network/Internet Consultants www.omnitec.net ============================================== -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
L. V. Lammert wrote:
On Tue, 10 Feb 2009, Per Jessen wrote:
L. V. Lammert wrote:
Rsync is much more friendly.
So is my cat, but what's that got to do with it?
Not all cats are friendly, .. but rsync always does a nice job.
No doubt about, it does - but comparing rsync and RAID1 is like comparing apples and oranges. /Per -- Per Jessen, Zürich (5.7°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Tuesday, 2009-02-10 at 16:20 +0100, Per Jessen wrote:
L. V. Lammert wrote:
On Tue, 10 Feb 2009, Per Jessen wrote:
L. V. Lammert wrote:
Rsync is much more friendly.
So is my cat, but what's that got to do with it?
Not all cats are friendly, .. but rsync always does a nice job.
No doubt about, it does - but comparing rsync and RAID1 is like comparing apples and oranges.
The point might be, that some people think that a raid setup protects you the same as a proper backup, and it doesn't. Regarding some types of problems (power failure, software failure, human failure, filesystem corruption) the protection of a raid is nil. It simply allows work to continue in the case of one disk failure. - -- Cheers, Carlos E. R. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.9 (GNU/Linux) iEYEARECAAYFAkmRn0MACgkQtTMYHG2NR9VG8ACfbq8+u2pVQDRS25SSzRhGjo37 Y0cAnjAncJQj/9ZnX3P2Y5I97CBJEoiq =+ECI -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Carlos E. R. wrote:
The point might be, that some people think that a raid setup protects you the same as a proper backup, and it doesn't.
They're two different things of course. One is about high availability, the other is, well, about backup. /Per -- Per Jessen, Zürich (5.3°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Carlos E. R. wrote:
The point might be, that some people think that a raid setup protects you the same as a proper backup, and it doesn't. Regarding some types of problems (power failure, software failure, human failure, filesystem corruption) the protection of a raid is nil.
It simply allows work to continue in the case of one disk failure.
-- Cheers, Carlos E. R.
If that was indeed the point, it was not evident from the post, where the user said they abandoned software raid in favor of rsync because the mistook rebuild time for unavailability time. No mention was made of backup, and in fact you are the first to mention backup in this entire thread. So it seems unlikely this WAS the point. Rsync does not guarantee backup either. A DELETION on the primary is just as likely as not to result in a deletion on the secondary. Using rsync or raid for backup is using like using a wrench to pound nails. But that has nothing to do with this thread. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
It sort of depends on the size of the disks. If this is the first sync it can take quite a while, but as long as one of the disks is empty, when the sync started there is no ambiguity as to which is more current, etc, and you can run right thru the sync. (That is you can still use the array while it is syncing without significantly extending the sync time). Those who make this out to be a problem really don't understand that this is meant to be a background process which does not hinder your active use of the array. Further, there is a full bandwidth option for the sync, especially useful for the first build. After that, use of the array is more important than getting them synced immediately. As Philip mentioned, one drive per channel in any given array is the rule of thumb with ATA/IDE or SCSI drives. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Is it really reasonable that it should take almost 48 hours to make
Anders Norrbring wrote: the initial sync of a 2-disk RAID-1 that is 200Gb in size?
The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
It sort of depends on the size of the disks.
If this is the first sync it can take quite a while, but as long as one of the disks is empty, when the sync started there is no ambiguity as to which is more current, etc, and you can run right thru the sync.
(That is you can still use the array while it is syncing without significantly extending the sync time).
Those who make this out to be a problem really don't understand that this is meant to be a background process which does not hinder your active use of the array.
Further, there is a full bandwidth option for the sync, especially useful for the first build. After that, use of the array is more important than getting them synced immediately.
As Philip mentioned, one drive per channel in any given array is the rule of thumb with ATA/IDE or SCSI drives.
Thanks to you all. Yes, naturally the 2 disks are on separate channels.. ;-) And I actually don't mind the initial sync time, I was just really surprised by the fact that it estimates 47.6 hours to make the sync.. As I mentioned, I've never set up a s/w RAID before, I've always been using h/w before.. Looking back at the old Compaq days with SCSI-2 drives, a RAID-5 with 12 disks very rarely took more than an hour to build, so 48 hours feels like "eternity" with 2 not-that-old disks. I need to look more into the mdadm utility, I actually just used the defaults to create the array, but of course there are tuning options, I just didn't expect that the result would be this.. ;-) It's really not a big issue for me, this is only a very temporary setup.. I was just surprised... Anders. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
Yes, naturally the 2 disks are on separate channels.. ;-) And I actually don't mind the initial sync time, I was just really surprised by the fact that it estimates 47.6 hours to make the sync..
I think that is _very_ unusual even though it's ATA133. To test it, I've just started a resync of a 500Gb array made up of two ATA133 drives - started at 0930, the initial estimate is 264minutes. (the system doesn't do anything else). /Per -- Per Jessen, Zürich (7.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
Yes, naturally the 2 disks are on separate channels.. ;-) And I actually don't mind the initial sync time, I was just really surprised by the fact that it estimates 47.6 hours to make the sync..
I think that is _very_ unusual even though it's ATA133. To test it, I've just started a resync of a 500Gb array made up of two ATA133 drives - started at 0930, the initial estimate is 264minutes. (the system doesn't do anything else).
/Per
-- Per Jessen, Zürich (7.4°C)
I guess the main issue here is that my host is quite busy running as a Samba server, and it also hosts a VMware virtual machine setup. Here's the info so far.. # cat /proc/mdstat Personalities : [raid1] md0 : active raid1 hdc1[1] hda1[0] 195358336 blocks [2/2] [UU] [=====>...............] resync = 25.8% (50513280/195358336) finish=1986.6min speed=1212K/sec Looks a bit ridiculous.. Maybe I should just lift off the disks and put them onto an idle machine, just to get it over with.. Current speed is hilarious.. ;-) Anders. (-12.9°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
I think that is _very_ unusual even though it's ATA133. To test it, I've just started a resync of a 500Gb array made up of two ATA133 drives - started at 0930, the initial estimate is 264minutes. (the system doesn't do anything else).
/Per
I guess the main issue here is that my host is quite busy running as a Samba server, and it also hosts a VMware virtual machine setup.
My resync has now been running a little over an hour, and is 22.8% done. Looks like the initial estimate is going to hold.
Looks a bit ridiculous.. Maybe I should just lift off the disks and put them onto an idle machine, just to get it over with.. Current speed is hilarious.. ;-)
It's not only ridiculous, it's dangerous. The longer it takes to resync, the higher the risk of the valid drive also failing. If the disks are the same make and model and were bought at the same time, the risk is even higher. /Per PS: -12.9C huh? It must be winter where you are :-) -- Per Jessen, Zürich (7.1°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
I think that is _very_ unusual even though it's ATA133. To test it, I've just started a resync of a 500Gb array made up of two ATA133 drives - started at 0930, the initial estimate is 264minutes. (the system doesn't do anything else).
/Per
I guess the main issue here is that my host is quite busy running as a Samba server, and it also hosts a VMware virtual machine setup.
My resync has now been running a little over an hour, and is 22.8% done. Looks like the initial estimate is going to hold.
Looks a bit ridiculous.. Maybe I should just lift off the disks and put them onto an idle machine, just to get it over with.. Current speed is hilarious.. ;-)
It's not only ridiculous, it's dangerous. The longer it takes to resync, the higher the risk of the valid drive also failing. If the disks are the same make and model and were bought at the same time, the risk is even higher.
/Per PS: -12.9C huh? It must be winter where you are :-)
Yep, it's winter in Sweden alright. Even the cat (no matter how friendly _he_ is) refuse to go outside today.. These drives are not of the same brand, nor the same size, they're only in the box to make a somewhat safe backup storage while moving from 9.3 to 11.1. So I won't use them when the upgrade is done, I just felt backing up some 160+GB would be a bit faster using drives than the old tape device in that box. Now I know better.. ;-) I'll pull the disks and put them in another machine which is idle, just to make the build and see how they perform in that one. I'll report back in a while.. Anders. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
I think that is _very_ unusual even though it's ATA133. To test it, I've just started a resync of a 500Gb array made up of two ATA133 drives - started at 0930, the initial estimate is 264minutes. (the system doesn't do anything else).
/Per
I guess the main issue here is that my host is quite busy running as a Samba server, and it also hosts a VMware virtual machine setup.
My resync has now been running a little over an hour, and is 22.8% done. Looks like the initial estimate is going to hold.
Looks a bit ridiculous.. Maybe I should just lift off the disks and put them onto an idle machine, just to get it over with.. Current speed is hilarious.. ;-)
It's not only ridiculous, it's dangerous. The longer it takes to resync, the higher the risk of the valid drive also failing. If the disks are the same make and model and were bought at the same time, the risk is even higher.
/Per PS: -12.9C huh? It must be winter where you are :-)
Yep, it's winter in Sweden alright. Even the cat (no matter how friendly _he_ is) refuse to go outside today..
These drives are not of the same brand, nor the same size, they're only in the box to make a somewhat safe backup storage while moving from 9.3 to 11.1. So I won't use them when the upgrade is done, I just felt backing up some 160+GB would be a bit faster using drives than the old tape device in that box. Now I know better.. ;-)
I'll pull the disks and put them in another machine which is idle, just to make the build and see how they perform in that one. I'll report back in a while..
Anders.
Now then.. I pulled the drives and hooked them up to another box, booting it up from DVD to rescue mode. I can see a _slight_ difference in rebuild performance, now it tells me it'll be finished in 18 minutes. From ~1000kB/s to ~68000kB/s. I guess that's it then. ;-) Anders. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Per Jessen wrote:
My resync has now been running a little over an hour, and is 22.8% done. Looks like the initial estimate is going to hold.
Forgot to keep an eye on it, but the resync finished at 14:02 - 500Gb in 4 1/2 hours. That's very, very close to the initial estimate. -- Per Jessen, Zürich (6.8°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Per Jessen wrote:
My resync has now been running a little over an hour, and is 22.8% done. Looks like the initial estimate is going to hold.
Forgot to keep an eye on it, but the resync finished at 14:02 - 500Gb in 4 1/2 hours. That's very, very close to the initial estimate.
-- Per Jessen, Zürich (6.8°C)
Same for me on the second box, it finished off the remaining 72% in 18 minutes instead of the 1900+ minutes on the busy machine. I'm starting to believe the ATA interfaces on the first machine isn't working as they're expected to, the difference doesn't seem to "be right" to only be related to system load. I re-installed the ready mirror set into the first machine and started a copy. That wasn't a very good thing it seemed. I got 2 md processes running at 100% cpu load after a couple of seconds, hogging everything, even the console got non-responsive for almost 20 seconds. Then I broke up the mirror and started the same copy onto one of the disks, no problems whatsoever. There's something fishy going on here, and I don't have a clue.. Annoying. Anders. (-8.5°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
I guess the main issue here is that my host is quite busy running as a Samba server, and it also hosts a VMware virtual machine setup.
Indeed, that will run the build at minimum speed, which is around 1MB/s (system default) You can tweak this, if you think your system can handle it: echo 10000 > /sys/block/md0/md/sync_speed_min should increase it to ~10MB/s Pit -- Dr. Peter "Pit" Suetterlin http://www.astro.su.se/~pit Institute for Solar Physics Tel.: +34 922 405 590 (Spain) P.Suetterlin@royac.iac.es +46 8 5537 8534 (Sweden) Peter.Suetterlin@astro.su.se -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Anders Norrbring wrote:
Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make
the initial sync of a 2-disk RAID-1 that is 200Gb in size?
The disks are of course a pair of old ATA-133 P-ATA disks, but it
still seems like 40 hours too many?
I have never made a software RAID before, only hardware, so I really
don't know..
Anders.
It sort of depends on the size of the disks.
If this is the first sync it can take quite a while, but as long as one of the disks is empty, when the sync started there is no ambiguity as to which is more current, etc, and you can run right thru the sync.
(That is you can still use the array while it is syncing without significantly extending the sync time).
Those who make this out to be a problem really don't understand that this is meant to be a background process which does not hinder your active use of the array.
Further, there is a full bandwidth option for the sync, especially useful for the first build. After that, use of the array is more important than getting them synced immediately.
As Philip mentioned, one drive per channel in any given array is the rule of thumb with ATA/IDE or SCSI drives.
Thanks to you all. Yes, naturally the 2 disks are on separate channels.. ;-) And I actually don't mind the initial sync time, I was just really surprised by the fact that it estimates 47.6 hours to make the sync.. As I mentioned, I've never set up a s/w RAID before, I've always been using h/w before.. Looking back at the old Compaq days with SCSI-2 drives, a RAID-5 with 12 disks very rarely took more than an hour to build, so 48 hours feels like "eternity" with 2 not-that-old disks.
You "feel" correctly... With new SATA 7.2K HD's a full RAID5 resync will take you less then 2 hours for each 400GB, if your system is manly idle. So for that amount of time you must have an hardware issue or a Controller/HD/driver incompatibility/BUG. There is a basic way to test you HD's speed ( not fully reliable but, should give you a hint ). Could you paste the output of 'hdparm -t /dev/<HDs>' ? Just remember to stop sync before you do it... Also, since your using PATA HD's, you should have better results by using libata. That means your HD's will "be seen" as /dev/sdX, just like a SATA/SCSI/SAS HD.
I need to look more into the mdadm utility, I actually just used the defaults to create the array, but of course there are tuning options, I just didn't expect that the result would be this.. ;-) It's really not a big issue for me, this is only a very temporary setup.. I was just surprised...
Anders.
-- Rui Santos http://www.ruisantos.com/ Veni, vidi, Linux! -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
John Andersen wrote:
Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
It sort of depends on the size of the disks.
If this is the first sync it can take quite a while, but as long as one of the disks is empty, when the sync started there is no ambiguity as to which is more current, etc, and you can run right thru the sync.
I'm not sure which one, but when an array is first assembled, one disk is assumed to be the valid copy, the other is assumed to be empty. -- Per Jessen, Zürich (7.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
John Andersen wrote:
Anders Norrbring wrote:
Is it really reasonable that it should take almost 48 hours to make the initial sync of a 2-disk RAID-1 that is 200Gb in size? The disks are of course a pair of old ATA-133 P-ATA disks, but it still seems like 40 hours too many? I have never made a software RAID before, only hardware, so I really don't know..
Anders.
It sort of depends on the size of the disks.
If this is the first sync it can take quite a while, but as long as one of the disks is empty, when the sync started there is no ambiguity as to which is more current, etc, and you can run right thru the sync.
I'm not sure which one, but when an array is first assembled, one disk is assumed to be the valid copy, the other is assumed to be empty.
True, and both my drives were written with zeroes all over.. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
participants (9)
-
Anders Norrbring
-
Carlos E. R.
-
Herbert Graeber
-
John Andersen
-
L. V. Lammert
-
Per Jessen
-
Philip Dowie
-
Pit Suetterlin
-
Rui Santos