> My more fundamental conclusion was that the whole master-master dream is a 
> myth:
> it doesn't solve anything. It doesn't scale at all.
> 
> Does it offer higher throughput?

NO

> Does it offer lower latencies?

NO

> Does it offer higher reliablity?

NO

> Does it offer higher availability?

YES!

With the multi-master database performance is no better unless you have
a massive read:write ratio. For most users I would say that this ratio
would be down around 1:1 (or for users who don't read their spambox this
could be lower reads to writes)

Remember that all writes have to be applied to both databases, so
performance on a busy site is going to be limited by this.

The reason I will shortly be switching to a multi-master setup is that
last night I spent a few hours after midnight fixing a problem with my
current setup (mysql files on a drbd store with heartbeat) - a simple
loss of network connectivity due to a switch that needed powercycling
caused both servers to reboot a number of times while fencing the data
store, and drbd failing to come up due to the network missing on the
second server, therefore the primary server was unable to get to the
mysql data.

Since multi-master mysql does not give any performance benefits over a
single instance (except maybe supporting a higher number of concurrent
connections), I will be using one server only at a time, and switching
to the secondary in a fail situation. I'll be using ucarp instead of
heartbeat as it's lighter and faster.

In this configuration there are no UID order issues, unless your system
continually fails. And a network failure to the secondary machine
shouldn't cause the primary to fail, as it did for me.

_______________________________________________
DBmail mailing list
[email protected]
https://mailman.fastxs.nl/mailman/listinfo/dbmail

Reply via email to