So Rackspace had troubles at one of its datacenters today. It took down a few major websites including TechCrunch. This got me thinking about how I would run a cloud service.
What I would do is have super high-end Internet pipes that would connect all my datacenters together. Each datacenter would have the exact same data. That means if one datacenter gets taken out, the others will be able to keep everything up. Kind of like server clustering. I would refer to this network of Internet pipes pumping huge amounts of bandwidth between all the datacenters as “SkyNet”. The service would be called “Brandon’s SkyNet”.
Who could resist a hot new cloud service called that? Eventually, “SkyNet” would be self-sustaining. Eventually, it could think and do whatever it wanted. I could then take lots of vacations while “SkyNet” did it’s thing. And no one’s blog hosted on “SkyNet” would ever go down!
Oh, wait…
Ok sarcasm aside, there has got to be a better way to do redundancy for a cloud service spanning multiple datacenters in different geological locations. Redundancy is key. If one datacenter goes down, something should kick in that keeps the services and websites up and running. While I was being funny about the whole “SkyNet” thing, I do think operating a service with datacenters as “nodes” isn’t such a bad idea. And when a note goes down, another one picks up the slack.
Tweet
Ok , not bad but I’ll see that and raise you one order of magnitude. Replace the node storage with routers that like bit torrent split the load . In other words encrypt and send storage to many multitudes of users whose computers become the cloud for themseles and for their neighbors and to participate we volunteer a portion and a partitition to the cloud.Like I said encyrption, and data splitting prevent privacy loss and each bit of data could be backed up on many computers.The benifit is of course is that failure of one or two machines is possible but not of a dozen. Of course the trick is the redundency of the router and encryption network. I feel that this could be backed up in real time across several parallel systems
TTFN Drew Schlais
I’ve been working in the online biz for around 3 years now. The big issues I’ve encountered have always revolved around power – either the price of it or the sudden and immediate lack of it. So I’d focus on power.
The generation of it starts with the usuals. I’ve been told of Scandinavian data centres that have their own wind turbines. They are located where there is a near constant supply of wind. Oversupply is sold into the grid. That cuts down costs nicely. It also allows for a 2-supply system – turbine failure -> pull power from the grid. In Iceland, we’re told the new data centres generate power using the natural volcanic heat from the ground. The same solution can be applied there. And both systems are carbon neutral – if you happen to believe or care in such things.
The multi replicating site approach (hmm, with our data centre being next toor to a certain one in Dublin, Ireland) sounds very familiar
It also allows the use of the lower cost container based approach. Less (almost no) aircon, modular systems, using less power and less real estate.
And if it is going to be called Brandon Skynet then it needs an army of droid operators – just to keep things going and nothing to do with world domination!