On Tue, 23 Aug 2011, Horacio Sanson wrote:
Thanks for all the answers... very informative and now I am confident
that what the client is asking is too much.
The application is a custom Ad Server (think adSense, iAd, etc) for
mobiles we implemented in Japan. And as some know the mobile market in
Japan is huge. We have three clients and between them we get
300,000,000 (300 million) hits per month. The image data changes
several times a day but the database and log files change hundreds of
times per second. Each record/line in the DB and logs represents lots
of money for advertisers and publishers that cannot be lost.
Our system is hosted in a cloud and they handle virtual machine
snapshots (they say) but I cannot tell how effective or good these
are. For example I understand these snapshots do not guarantee that
the DB is in a stable because they cannot flush the DB memory during
the snapshot and the same may be true for logs.
As many said this 15 min requirement is too much for any backup
solution and instead I will go with high availability options like
redundant DBs (slaves) and remote logging (rsyslog).
thanks for all the answers, very helpful....
regards,
Horacio
If this helps, I setup database replication here and then am doing
backup from one of the slaves. This way I do not bother the master with
this mucking around with backups. there are more clever solutions but I am
lazy. ;)
