Melvin Lancelot's Blog

Practically every one with an online business model is now referring to their service as cloud computing - starting from your average Joe hosting firm all the way to the SaaS/S+S vendors, every wants to ride the next buzzword wave, and this distorts the cloud computing term all together. Part of the reason being that historically the term "cloud" loosely referred to anything that’s available online/on the Internet. Ask a bunch of geeks and you would get a different explanation of cloud computing from each person (if you ever asked a bunch of people what Web 2.0 is you know what I mean). So how can we qualify if a service is really leveraging the cloud computing model?

That’s a tough question. An easier way to answer this is by first examining the behavior of services provided by some of the well known cloud computing vendors. Lets take Amazon and Google as two examples, both have different business models but under the hood when we use their cloud computing service within our application/service the 3 common behavior that we see from it are:

a. Scalability
b. Availability
c. Economical/cost effective

Lets discuss scalability first. Imagine you're tasked to design an online application/service that should be "Internet scalable". Imagine (if you will) that you're designing the next big social networking/ the next big YouTube etc. How do you go about designing it to support millions of users? For that matter how does anyone do it?

The short answer - you do that iteratively. Iteratively is a good euphemism, the reality is that you do that after several design blunders and limitations :). In the iterative approach you would first design a cost effective solution to scale for a smaller audience and then when you start seeing more traffic you add more hardware till the point it doesn’t improve anything then you start to redesign for better scalability. This is how Amazon and Google have grown as well and have designed their overall system to support such high scalability - but interestingly they have abstracted their design to a degree that it could be repackaged into a subscription-based service - a cloud computing service.

But first how do you build a massively scalable solution? How would you build your architecture to accommodate linear scalability?

Any architecture can be described as being composed of two types of components:

a. Stateless components
b. State-full components

Stateless components are those that only do some processing on data and don’t persist state - hence are easily to scale via a scaled out design which as a byproduct also gives you higher availability, also using scaled out architecture you can keep adding inexpensive boxes to the system thereby reducing your cost while the system continue to keep humming.

Statefull components on the other hand are those that persist state of resources that it need to work with - for e.g., File system, databases, BLOBs etc. traditionally the only option to scale these would be to via a scaled-up approach - i.e. beef up the hardware on the server. This is more expensive than a scaled out architecture and introduces single point of failure. Traditionally we have been using approaches such as data replication etc to scale out these components but it introduces several complications. Fundamentally statefull components are the ones that impact overall performance and scalability of a system.

Statefull components may be notoriously difficult to scale, however even stateless components can present scalability challenges when your planning for massively scalable/internet scalability scenarios. Cloud computing vendors have made significant investments in technology to ensure that compute intensive processing can been parallelized and distributed to the max. Googles MapReduce is a elegant approach to solving these challenges.

We need a new breed of products to handle how stateless and statefull components can be scaled and distributed, the traditional approach of persisting state in the database etc just doesn’t make sense, increasingly we see that the trend is not to store atomic level transaction information in a database as we used to but store it in the form of blobs, but at the same time ensure that we do so while managing resources economically. Large part of Amazons and Googles innovation (their magic sauce if you will) in cloud computing involves developing these proprietary components.

Another way of looking at the scalability that cloud computing gives you is the ability to scale your computing resources as an when you see demand - after all it’s a pay as you go model. Traditionally online companies have been provisioning for hardware to meet spikes in traffic (like holiday seasons etc). That would also imply that they are unnecessarily paying for the extra scale which they don’t leverage all the time. By hosting it via a cloud vendor they can dramatically reduce their operational cost.

Well its not just Amazon and Google that are thinking in this direction, several platform vendors like Microsoft and other open source groups are releasing products are that will address these challenges. You can expect to see some radically different products being released from these vendors that address the distributed massively-scalable challenges. You can also expect (hosting)companies to leverage these prepackaged cloud computing capabilities and provide it as a subscription service.

Cloud computing vendors are able to provide Internet scalability at an affordable cost and can potentially give you a better SLA that if you were to manage your own infrastructure - that’s the overall package that makes cloud computing so compelling, probably best described by Jef Bezos - "You don't generate your own electricity, why generate your own computing?". Arguably there are other factors that influence your vendor decision but we hope that the next time your evaluating a cloud computing vendor/solution or building your own, you know what to look for.