People who work in IT joke about putting fog machines in their datacenters so that they can legitimately call it “the cloud”. Here’s why that’s funny: The cloud is a marketing term. The term has no universal meaning. The general consensus right now is that it’s a remote datacenter, often with some type of distributed computing or redundancy baked in. With the portability introduced by virtualization, it’s possible to have servers that work the same way even as they move between physical locations, so that the location of the data becomes less relevant.
There are many, many different types of clouds, but we can divide them into two basic groups.
Public Cloud – when people say this, they’re usually referring to hosted services like salesforce.com, Google Apps, or Office 365, but this can also mean hosted server instances provided by Microsoft, Amazon, or Google. The key is that it’s infrastructure that’s shared by lots of people, and it might be far away.
Private Cloud – usually this refers to something that has been purpose built for an organization, and while it’s in a remote datacenter (as opposed to “on premise” – in the customer’s office) it’s usually in a location that is somewhat accessible, on infrastructure that’s either not shared or that’s shared between a select few organizations, that’s geographically desirable (distance often introduces latency because traffic travels across more ‘hops’ before reaching its destination) and is usually managed by a company’s internal IT team or Managed Service Provider.
So which is right for you?
Public cloud probably got its start when Amazon started selling their unused computing power. You see, there’s one day when Amazon needs to have lots of servers to be able to run its retail business, and it’s called Black Friday. They have to own enough infrastructure to be available for the sales the day after Thanksgiving. Because of this, they have a lot of unused infrastructure for about 363 days out of the year. They started selling this capacity to other companies, and in doing so, they turned computing into a commodity and made the “mainframe” model of computing attractive again. Since they had fixed costs, they made it really cheap to use their stuff. Google and Microsoft (Azure) followed, then Rackspace and Softlayer (IBM) and a few dozen others, but Amazon’s necessary scale and fixed cost have made them the dominant player in the public cloud space. On one hand, the public cloud is very much a “nickel and dime” environment, which makes it hard to resell, because they charge you for data transfer, they charge you for cpu utilization, they charge you for disk storage, etc. It’s hard to predict what it will cost to do a public implementation, but it’s usually cheap. Amazon has cut their cloud pricing over 40 times since 2008.
The best thing about public cloud is probably that it doesn’t really require any hardware literacy, or any efficiency. Developers can write woefully inefficient applications, and just keep throwing more power at it until it works. If a software service becomes popular, they can just keep adding clusters and processors until it’s stable, and they can do this with mouse clicks on a reasonably easy to navigate website.
Private cloud thrives in circumstances where regionalization is a factor, for example a company called US Signal sells a private cloud product where you get mirrored servers in two different regions of Michigan. Anyone in the world can buy it, but it’s especially attractive if you’re doing business in Michigan, because you’re probably making it to that data in just a couple of hops, and you can bake in MPLS and other technologies to get latency almost equivalent with what you’d get with an on premise server. You can potentially save some money this way, and costs are a bit more predictable, but this model requires more maintenance and management. We partner with a local datacenter that buys lots of fiber from Cox Communications, which is the dominant ISP in south Orange County. Consequently, for our customers in California we can outperform a lot of what Amazon has to offer, because their big west coast datacenter is in Oregon. For this reason we joke about calling it “smog” because it is closer than the cloud.
In either public or private settings, there are some big gains that come into play when a business moves its data out of the closet and into a cloud. First, scalability is simpler. You can add more infrastructure faster. Second, managed datacenters are much more resilient. They have physical security, backup generators, climate control, up to date disaster recovery plans, and a series of other safeguards to protect productivity.
Some people like the private model because it’s somewhat more controlled, and less discoverable by attorneys and governments. Citing privacy concerns, some people don’t trust Amazon, Google, and Microsoft to tell them if their data has been accessed under subpoena or warrants. Although we like private cloud for other reasons, if the NSA wants your data, they are probably going to get it anyway, and the larger providers are at least following best practices surrounding security, which is a full time job these days. What’s probably a more legitimate objection is that these large datacenters could become targets for denial of service, physical disruption by acts of terrorism, and other behavior based events that can cause high demand, such as Black Friday shoppers. Shared infrastructure is more efficient for the vendor, but if your neighbors are getting hit with viral traffic, so are you.
So hopefully that demystifies “the cloud” a little bit. The most important thing to understand is that the data is “somewhere else”. We love this stuff, so let us know if you want to talk about the weather.