If you think it costs 3X to be in three availability zones then you do
need better engineers...
We run a third of our systems in each zone normally, to avoid the bad
zone we moved to be running half our systems in each of two zones.
About the same number of systems total, and the cost of the systems is
the dominant cost.
Since the cloud is elastic (remember, that was the point :-) we don't
need to pre-allocate capacity in each zone to take the extra load,
unlike a datacenter DR solution.
The cost of running in multiple zones is a minor increase in network
cost, slightly more latency, and that you have to decide to do it up
front. This works for any scale, nothing to do with running large
scale. If you don't have enough instances to spread three ways, use
On Apr 24, 10:08 am, Miha Ahronovitz <mij...@sbcglobal.net> wrote:
> On 4/23/2011 5:31 PM, kowsik wrote:> Netflix, for
> > example, runs in 3 different regions (see @adrianco's tweets).
> Sure, this a good thing , but how much this costs? Netflix footprint
> on AWS is huge and cost a huge X (Netflix repeatedly refused to disclose
> how much they pay AWS for hosting their Data Center at each meetup
> gathering where their top technical people spoke)
> Now if they host on 3 regions, quite simply this means a cost is 3X
> versus 1X. Say they got a deal. They still pay , say 2X, and it should
> be two HUGE Xs.
> Netflix has a group of rock stars that claim credit for the success of
> Netflix port to AWS and now their survival of AWS failure. But not every
> company has the same capabilities. And even if they have the same
> capabilities (like Zencoder and Quora, both with superb engineering
> talent) they do not have the same $ to spend.
> The Zecoder blog entry is must readhttp://blog.zencoder.com/2011/04/22/skynet-ec2-and-zencoder/ >
> As a trainee engineer, I was told the following definition of an
> engineer: "An engineer is someone who can build with $1M the same
> thing that any fool can with $100M"
> < 1KViewDownload