Slashdot Mirror
The Risk of a Meltdown In the Cloud
zrbyte writes "A growing number of complexity theorists are beginning to recognize some potential problems with cloud computing. The growing consensus is that bizarre and unpredictable behavior often emerges in systems made up of 'networks of networks,' such as a business using the computational resources of a cloud provider. Bryan Ford at Yale University in New Haven says the full risks of the migration to the cloud have yet to be properly explored. He points out that complex systems can fail in many unexpected ways, and he outlines various simple scenarios in which a cloud could come unstuck."
I don't understand the intent of the article other than to provide a knee-jerk chicken-little response to cloud processing and storage.
Not one of the items mentioned is unique to the cloud. It can happen to any data center with more than two nodes involved in a cluster.
But that's not surprising, because "the cloud" is just a distributed collection of cluster servers, the same as large multi-nationals have been running pretty much since their customer loads exceeded the ability of one server to span the global community.
I do not fail; I succeed at finding out what does not work.
I think the safest bet is to have local copies in addition to copies in the cloud, even if all the processing and computing is actually done in the cloud. Companies should set stuff up to keep a local copy of critical services on a good old fashioned tape drive or backup server. This is sort of a reverse of the cloud based backup solutions, where local processing and databases took place on local servers, but had backups in the cloud in case of a local disaster. Same idea: Have a local backup in case of the meltdown of the cloud. You may find your primary app is temporarily useless, but you at least have all your critical data (hopefully in a format that can be transferred.)
Occasionally living proof of the Ballmer peak.
The general point is that it's possible to get bad emergent behavior which is unexpected. This shouldn't be surprising to anyone (but it is, alas). We see it over and over in complex systems, and it's got pretty much nothing to do with what you implement the complex system with.
What to do about it? Well, the only real fix is to stop the drive for efficiency at all costs. All those little inefficiencies that hit your bottom line, they also mean that when things go wrong you can weather the storm more easily. And yes, that resilience means things are going to cost more. How much more? Well, depends how much risk you want to take out of the system and how much you're willing to pay. Your call. (A local backup removes a lot of risk from things like cloud providers going belly up unexpectedly, but it does mean you're stuck with actually having to pay real money to do the backup and make sure it is working.)
"Little does he know, but there is no 'I' in 'Idiot'!"
If someone loses "all" their data in the cloud, their problem has nothing to do with the cloud. If you lose all your data, it's because you kept all your data in one place, with no backups in a different place, and all fault lands on you, not the cloud, not your cloud provider, and not on any given piece of technology. There have already been large failures, and some companies have already lost massive amounts of data, but it doesn't change anything, because these problems have nothing to do with whether you host your own servers or rent them from somebody else, which is really all "the cloud" boils down to.
Also, inherently not trusting virtualization as a concept in 2012 is is moronic and baseless. It's a technology, just like any other. It can be implemented well or it can be implemented poorly, but as a concept it is not novel or revolutionary to any degree that it should engender trust or distrust, at least not any more than the hardware underneath it.
If the cloud is not more robust than what your grandma could come up with on her own then what's the point really?
Isn't the whole point of "the cloud" the fact that you aren't managing this stuff yourself? You don't have the burden? You don't need the expertise?
If you push it back on the cloud consumer then a lot of it is really quite pointless.
A Pirate and a Puritan look the same on a balance sheet.
"If the cloud is not more robust than what your grandma could come up with on her own then what's the point really?"
Money.
I am John Hurt.
I keep remembering the Google applications that keep disappearing. Data is one thing, application providers going out of business or discontinuing a service is another issue entirely. Hopefully the competition still stands and you can migrate your applications over to their service.
I object to power without constructive purpose. --Spock
Just a thought. Forget actual failure. What happens when they have IP data or licensed data that is being hosted by a cloud provider, or company to company lawsuit. Court case starts. Could or would they they seize all computers / servers that could house the data? What would happen to the other peoples data that resides on the same physical hardware?
love the taste, hate the texture
Pardon me mods, but +4 informative? This is a terrible summary from someone who doesn't seem to have understood what he's read. The novel "cloudy-thing" aspect of the article's argument is the very part the parent misses when he dismisses this as "nothing new".
The cloud is an abstraction that intentionally hides detail. Cloud providers do that to make the service being offered simple to package, sell and use. They also do what they can to keep the tricks of their trade secret from competetors. But their infrastructure is actually very complex relative to what the average small to medium client would need for themselves. This is important in three ways:
This is an important and novel issue and worthy of some real consideration.
Yes, a lot of people lose data on mainframes and the problem was generally behind the console. Online production systems and job-based systems need to be designed with practicality and component failure in mind.
The cloud is a system, and the system needs redundancy, checks and reality checks, and quality ins-and-outs. That's right, just like what you've been doing all along. Same security, same backups, same contingencies-- as there are no shortcuts, just cheaper hardware.
I have to dismiss the cited article as it's entirely ephemeral, with not one single citation to back it up. Financial markets, while important, are a somewhat unique context to cite; they're built differently than many cloud components. There is nothing tangible cited, no case history, just some Yale-y's bad, after-lunch tummy growling coinciding with thinking about the complexity of modern infrastructure.
This whackamole approach serves no purpose, except to sell more inside, rather than outside, infrastructure.
---- Teach Peace. It's Cheaper Than War.