On 08/15/2010 09:49 AM, nux@xxxxxxxxx wrote:
> Tuptus writes:
>
>
>> Nic by się nie stało gdyby Agnello George nie napisał:
>>
>>> we have multiple servers approx 10 and each has about 100 GB of data in
>>> the /var/lib/mysql dir , excluding tar , mysqldump and replication how do
>>> we
>>> take backup for these databases on to a remote machine and store them
>>> datewise , ( the remote machine is a 2TB HDD )
>>>
>>> currently tar is not feasible as the data is too huge and the same goes
>>> with mysqldump
>>>
>>> suggestion will be of great help
>>>
>> Why not mysqldump?
>> I suggest mysqldump to local dysk and backup this to remote.
>> I use it with __bacula__.
>>
>> --
>> Tuptus
>>
> AFAIK mysqldump locks the tables.. and to have the tables locked while you
> dump 100 GB of data is very annoying, if not unacceptable.
> The best solution by far in this circumstance (yeah, i know he said he
> doesn't want replication) is to have a master/slave replication and perform
> the dump on the slave.
>
No, the best solution in this circumstance is to use a master/slave
replication and use LVM to take a static snapshot of the slave. That
lets you use rsync efficiently to then sync it to the remote box.
A straight dump doesn't work efficiently because there isn't a good way
to 'null out' unchanged data from being copied in a regular dump during
the remote sync. With an LVM snapshot the file structure is preserved
between runs allowing rsync to only send *changes* instead *everything*
after the first sync. It could easily be the difference between taking a
day to do the remote sync and taking 1 to 2 hours to do it (assuming you
can read from your drives at around 40 MBytes/sec sustained but can only
get 10 mbits/second sustained over the network).
--
Benjamin Franz
_______________________________________________
CentOS mailing list
CentOS@xxxxxxxxxx
http://lists.centos.org/mailman/listinfo/centos
