[Gllug] Geographcial server failover

[Andrew Farnsworth]

Doug Winter wrote:
<snip>

> Just echoing Tet really - everyone thinks they need 100% uptime till 
> they see the price :)
>
> In real life, very few places need that sort of availability.
>
> I'm also generally of the opinion that the massive additional 
> complexity of systems that provide for multi-master geographically 
> distributed transaction systems causes more downtime than it solves.

It certainly causes more headaches :-)  One thing many people don't 
really realize is what statements of uptime really mean, here is a quick 
table:


	Min / year 	Min / Month
99% 	5256 	438
99.9% 	525.6 	43.8
99.99% 	52.56 	4.38
99.999% 	5.256 	0.438


The Famous "Five 9s" of service still mean 5 minutes of down time per 
year.  Upon reflection, most people can handle 5 minutes / month or even 
45 minute / month... reaching Three 9s of uptime isn't really that hard 
to achieve, good components will usually do the trick along with a good 
Colo facility.  Getting Five 9s or better is more than my money is 
worth.  The very few situations I have seen where down time is a real 
issue is with Stock trading where five minutes down in the middle of the 
day can cost you millions and even a small hiccup can cost tens of 
thousands.

Andrew Farnsworth



-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug