Prev: Shared Memory and mutex protection
Next: Simple Hack To Get $2500 To Your PayPal Account.
From: Bin Chen on 21 Jul 2010 01:36
Hi,

I am doing some fail-over research, there is one question comes up to
me. If the server1 is down unexpectedly, how can the data be migrated
to server2? I think this question is not a YES/NO question, there must
be some mechanism that can make the data loss as little as possible.
Using pure software, is there any good framework/paper existing there
for me to refer to?

Thanks.
Bin
From: Daniel Molina Wegener on 21 Jul 2010 07:35
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

On Mi� 21 Jul 2010 01:36,
Bin Chen wrote:

> Hi,
>
> I am doing some fail-over research, there is one question comes up to
> me. If the server1 is down unexpectedly, how can the data be migrated
> to server2? I think this question is not a YES/NO question, there must
> be some mechanism that can make the data loss as little as possible.
> Using pure software, is there any good framework/paper existing there
> for me to refer to?

You must search for "replication" as keyword. There are some existing
frameworks and software tools. Also it depends on what you need to be
replicated. For example databases can work with master/salve and
master/master models, similar models can be applied to filesystems.

If you are searching for database replication, search for master/slave
and master/master models, and for file systems, just search for replication
services. Replication models for databases are dependent of the engine
that you are using, and file system replication services are dependent
on the software that you are using. For example for Linux systems, I've
found:

http://sourceforge.net/projects/afrs/

If see the description, it uses the inotify(7) facility, which looks
for changes on file systems and directories with a behaviour very similar
to real-time file system monitoring.

>
> Thanks.
> Bin

Best regards,
- --
Daniel Molina Wegener <dmw [at] coder [dot] cl>
System Programmer & Web Developer
Phone: +56 (2) 979-0277 | Blog: http://coder.cl/
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
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=WunQ
-----END PGP SIGNATURE-----

From: Scott Lurndal on 21 Jul 2010 11:11
Bin Chen <binary.chen(a)gmail.com> writes:
>Hi,
>
>I am doing some fail-over research, there is one question comes up to
>me. If the server1 is down unexpectedly, how can the data be migrated
>to server2? I think this question is not a YES/NO question, there must
>be some mechanism that can make the data loss as little as possible.

Best way is to move the storage off the server via fiberchannel, iSCSI or
a NAS protocol. Let the storage subsystem handle the data (e.g. raid5).
Zone the lun to the primary and backup servers. Ensure the software
can roll-back any partially completed transactions at the time of the
failover.

>Using pure software, is there any good framework/paper existing there
>for me to refer to?

DB2, Oracle, Sybase all support this type of failover with data consistency;
I'm sure they all have white papers that would be helpful to you.

scott
From: David Schwartz on 22 Jul 2010 08:24
On Jul 20, 10:36 pm, Bin Chen <binary.c...(a)gmail.com> wrote:

> I am doing some fail-over research, there is one question comes up to
> me. If the server1 is down unexpectedly, how can the data be migrated
> to server2? I think this question is not a YES/NO question, there must
> be some mechanism that can make the data loss as little as possible.
> Using pure software, is there any good framework/paper existing there
> for me to refer to?

If you need to failover data, then that data should not be stored on
just one system. Your data flow goes like this:

1) System gets data.

2) System replicates data to failover system.

3) System processes data itself.

4) System acknowledges data.

The order of steps 2 and 3 may be reversed depending on circumstances.
The data has to be replicated *before* the failure.

DS
From: mac on 24 Jul 2010 13:04
> ...If the server1 is down unexpectedly, how can the data be migrated

Two possibilities are
1) the data is connected to more than one server.
2) you have (always) already migrated eniugh information to reconstruct
the data

For 1) use dual-port devices and busses
For 2) make a copy and send along any changes. You may not need to
failover all changes, only those before some kind of checkpoint or those
not made in the last 30 seconds.
--
mac the naïf
 | 
Pages: 1
Prev: Shared Memory and mutex protection
Next: Simple Hack To Get $2500 To Your PayPal Account.