Please explain "cloud computing" to me?

[Learjeff]

@deltasigma

I don’t really know much about how it’s implemented, but the impression I get is that “the cloud” really falls into 2 main categories, software services and storage.

Three main ones, actually:
storage
application
system

You might classify application and system as the same, but technically they’re handled very differently. For application, you just run the application. For system, you appear to have hardware.

And a new one is cropping up, as I mentioned above: communications (virtual networks). That’s really none of the above. Simplest case is having multiple virtual servers connected by some virtual network.

For the [storage], I don’t really understand the security issues since all you really need to do is encrypt the data.

You have to trust the provider if the provider is providing the encryption. You also have to trust the provider not to simply lose the data.

And even for software services, since those can run in dynamically created virtual machines, I still don’t see why security should be an issue assuming that you can verify that the the VM is secure.

How do you verify that the VM is secure? Its inner workings are available to the service provider. In any case, those who want to use application services don’t want to manage a VM.

I also see no one has mentioned distributed computing. I volunteer time on several of my personal machines including several fairly high end dual processor servers for medical research projects. They insure the validity of the work done on their project by having a quorum for each work unit. IOW, ever WU gets sent out 2 or more different client machines for processing. The results returned have to be validated and match or the result is discarded and the WU is reprocessed.

Using this method the project emulates a supercomputer with power something of the order of few petaflops.

Right, but it’s not quite “the cloud”. Both are cases of distributed computing. Cloud-computing is a refined subset of distributed computing. There are certainly a lot of common elements to the two cases, but they’re not the same thing.

Cloud computing requires distributed computing, but not vice-versa. Furthermore, cloud computing uses a set of standards and mechanisms that are similar to but not the same as the frameworks used to implement massively parallel solutions. (I use the term “framework” generically here, and not in its software jargon meaning.)

[Voyager]

@Learjeff

Right, but it’s not quite “the cloud”. Both are cases of distributed computing. Cloud-computing is a refined subset of distributed computing. There are certainly a lot of common elements to the two cases, but they’re not the same thing.

Cloud computing requires distributed computing, but not vice-versa. Furthermore, cloud computing uses a set of standards and mechanisms that are similar to but not the same as the frameworks used to implement massively parallel solutions. (I use the term “framework” generically here, and not in its software jargon meaning.)

Distributed computing usually involves splitting task into pieces and sending them to different computers, like in deltasigma’s example. The pieces are usually not totally independent. For the most part all the jobs sent to the cloud are totally independent, and so you don’t have the same scaling problems seen in distributed computing. The vast majority of jobs run on our computing ranch are in fact totally independent, though you can do distributed computing on it if you wish.

[leahcim]

@Learjeff

Cloud computing requires distributed computing, …

I disagree with this statement, in that there is nothing that requires that a cloud service have any sort of distributed aspect (unless you count remote log-in as a “distributed”). AWS micro instances are exactly the opposite of that – small servers without a lot of memory or processing power ideal for things like a low-capacity web server that doesn’t require much to work properly.

Conversely, you can run a hadoop cluster using a local rack of servers just fine. It’s expensive, but for the security-conscious it’s ideal.

Of course, the big cloud providers, Amazon and Google, have a business model of “we know how to make big distributed data systems very well as part of our core business, so pay us to make yours instead of making your own” so their offerings are going to be more weighted to the big stuff, but there’s nothing that makes a provider solely of small instances not “cloud computing”.

[deltasigma]

I’m not sure what you mean by “independent”, but if you mean that the result of one determines the input to another, that’s not the case - or at least that’s what they have told me for most of the projects I’ve been involved in.

That’s a necessity since these work units have to be created and uploaded weeks if not months in advance by the project scientists so any dependencies of that nature wouldn’t be practicable if even possible. We have however run projects in different phases where, and this is just assumption, the results of the prior phase guided the creation of work units for the following phase, but that’s really not the same thing.

However I don’t really get involved in the actual mechanics. I’ve just been involved with the project for so long I’ve picked up a few things here and there.

In terms of secure VM’s, I was under the impression that there were versions of Linux that were developed specifically for this purpose - TAILS for example. I suppose that you try to incept the machine’s state at various points in time and spy on it in other ways, but it also seems that there ought to be a way to at least partially encrypt some of the operations. Of course anyone who has control of the physical hardware has the advantage, but you should at least be able to make it a sufficiently onerous process that only very determined people with sufficient resources would make the attempt - in which case you probably shouldn’t be in the cloud anyway.

edit: oh, in terms of losing data, I use the cloud for a variety of purposes but I always keep a local back up. A couple actually.