| Author |
Message |
Cooler
Guest
|
 Posted:
Sat Sep 24, 2005 12:16 am Post subject:
Mirrored volume - data immediately available on the mirrored |
|
|
Hello all,
I have a customer who is adamant that there is some software or
hardware or combination of both that allows the secondary volume of a
mirror to instantaneously see the data that was created on the primary.
In all of my experience with both Windows and UNIX environments I have
not come across anything that does this.
Of course, the mirror can be created at the hardware level and in that
sense it is an exact mirror. However, in order for the filesystem to be
able to pick up the synced data it must be unmounted and then
remounted, usually with an fsck on UNIX prior to remounting.
This applies to whether or not the mirrored volume is on the same host
or a different host altogether.
Can anyone dispute this, if so can you tell me how the software is able
to pick up the changes to the volume and push it up to the filesystem
level dynamically?
Also, can anyone give me a good explanation of why this isn't possible
if this turns out to be the case? I believe it has to do with the
filesystem needing to reread in its inode table but can't say for sure.
Thanks for any advice. |
|
| Back to top |
|
 |
Rob Turk
Guest
|
| Posted:
Sat Sep 24, 2005 12:16 am Post subject:
Re: Mirrored volume - data immediately available on the mirr |
|
|
"Cooler" <keithcp@gmail.com> wrote in message
news:1127506426.880106.103300@o13g2000cwo.googlegroups.com...
| Quote: | Hello all,
I have a customer who is adamant that there is some software or
hardware or combination of both that allows the secondary volume of a
mirror to instantaneously see the data that was created on the primary.
In all of my experience with both Windows and UNIX environments I have
not come across anything that does this.
Of course, the mirror can be created at the hardware level and in that
sense it is an exact mirror. However, in order for the filesystem to be
able to pick up the synced data it must be unmounted and then
remounted, usually with an fsck on UNIX prior to remounting.
This applies to whether or not the mirrored volume is on the same host
or a different host altogether.
Can anyone dispute this, if so can you tell me how the software is able
to pick up the changes to the volume and push it up to the filesystem
level dynamically?
Also, can anyone give me a good explanation of why this isn't possible
if this turns out to be the case? I believe it has to do with the
filesystem needing to reread in its inode table but can't say for sure.
Thanks for any advice.
|
I think what your customer is referring to is a snapshot. This is a
secondary view on a filesystem, which is initially identical to the main or
primary volume. Handy for instant restores, but also for backups of a
'frozen' consistent filesystem while the primary copy continues to be in
use. Implementations exist both in hardware and software.
Either that, or your customer refers to synchronous data replication. In
that case any write actions are sent to both the primary and the secondary
file system, and only after both have completed the write, it will be
acknowledged to the host. Usually this functionality exists on midrange and
high-end fibre channel disk systems. Some appliances can offer similar
functionality (FalconStor, SANrad, Datacore, just to name a few). Some do
the same over IP networks.
Rob |
|
| Back to top |
|
|
Cooler
Guest
|
| Posted:
Sun Sep 25, 2005 8:15 am Post subject:
Re: Mirrored volume - data immediately available on the mirr |
|
|
You're correct Rob, I am referring to snapshots, specifically
copy-on-write.
Can a secondary host see the snapshot data on the secondary volume
without unmounting and remounting the drive/volume/filesystem?
Thanks. |
|
| Back to top |
|
|
Guest
|
| Posted:
Thu Oct 06, 2005 8:16 am Post subject:
Re: Mirrored volume - data immediately available on the mirr |
|
|
Your question basically is, will an operating system handle properly if
you go and change the data on a mounted file system without it knowing
about it. And the answer is, it depends. If you disable all caching it
will probably work. However without unmounting and remounting, it may
have some data cached and wont bother to look to the disk to see if has
changed. So it will work, but the OS may now know about it.
Does that make sense. Technically what your customer is saying can be
done, very easily, but in real life you wouldn't do it in fear of
data corruption on the secondary site or the OS not knowing about it. |
|
| Back to top |
|
|
Anton Rang
Guest
|
| Posted:
Mon Oct 10, 2005 4:16 pm Post subject:
Re: Mirrored volume - data immediately available on the mirr |
|
|
"Cooler" <keithcp@gmail.com> writes:
| Quote: | I have a customer who is adamant that there is some software or
hardware or combination of both that allows the secondary volume of a
mirror to instantaneously see the data that was created on the primary.
In all of my experience with both Windows and UNIX environments I have
not come across anything that does this.
|
The SAM-QFS file system on Solaris can do this.
| Quote: | Can anyone dispute this, if so can you tell me how the software is able
to pick up the changes to the volume and push it up to the filesystem
level dynamically?
|
When a file is accessed, its inode is verified against the copy on
disk (if a certain time period has elapsed). If the copy on disk has
changed, the inode in memory is updated and any cached pages are
flushed. Directories are handled similarly.
This isn't really an entirely general-purpose solution, since it's
possible to see an "inconsistent" state of a file in the general case
(e.g. a partially completed write). It works well in controlled
environments, though. A typical use is for a web farm, where updates
are pushed by a single machine to a bank of servers. By performing
the updates "atomically" using rename, the servers always see a
consistent image for any file.
| Quote: | Also, can anyone give me a good explanation of why this isn't possible
if this turns out to be the case? I believe it has to do with the
filesystem needing to reread in its inode table but can't say for sure.
|
It's certainly possible but the file system has to be designed for it.
In general this means partially disabling their cache (or at least
adding validation). Since this hurts normal performance, file systems
don't usually do it. :-) For SAM-QFS, it's a mount option ("reader"
or "writer"); readers are forced to validate their caches after a
timeout period, writers are forced to flush any changes to disk more
quickly than usual.
-- Anton |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in