Cloud Computing – Where does one start from

Consider this analogy: historically, a developer determines when it’s appropriate to create new threads. Today, a developer can use an API to attach a new server with the same ease. The ability to program an application architecture dynamically puts enormous power in the hands of developers with a commensurate amount of responsibility. When this power is well understood and harnessed, the results can be spectacular. Animoto’s mashup tool that creates a video from a set of images and music scaled from 50 to 3,500 servers in just three days due in part to an architecture that allowed it to scale easily. For this to work, the application had to be built to be horizontal scaled, have limited state, and manage its own deployment through APIs.

Similarly, when the New York Times wished to convert 11 million articles and images in its archive to PDF format, one developer using 100 Amazon EC2 simple Web service interface instances running Hadoop completed the job in 24 hours for less than $300 of software development. The potential is immense. But if an application has a lot of state and can’t be deployed programmatically, it probably isn’t an application that is suited to this model.

Having standards that make deployment easy is more important than having the perfect environment for the job. There will continue to be specialization, however the 80/20 rule applies and the starting point should be with a standard.

Loose-coupled, Stateless, Fail-in-place computing

Cloud computing assumes that everything is temporary, and it’s just as easy to redeploy an entire application then it is to manually patch a set of individual virtual machines. If data is small, or the application is stateless, a hybrid cloud can be much more useful than if large amounts of data must be transferred into a public cloud for a small amount of processing. The classic example of Cloud computing with parallelisation with load balancing is a number of stateless Web server all accessing the same data, where the inkling workload is distributed across the pool of servers.

The cloud computing philosophy for application maintenance is not to patch, but redeploy. Managing the model that created a virtual machine image, not the image itself, simplifies this redeployment. Application images are not patched; they are throwaway objects and thus need to be stateless. A component should be able to “fail-in-place” with little or no impact on the application. This requires a paradigm shift for administrators and developers.

Service Offering Types

Customers can manipulate not just virtual machine images, but also servers, storage systems, network devices, and network topology. Hybrid clouds combine both public and private cloud models, and they can be particularly effective when both types of cloud are located in the same facility. Cloud service providers tend to offer services that can be grouped into three categories: Software as a Service, Platform as a Service, and Infrastructure as a Service. The first and ‘oldest’ is ‘Software as a Service (SaaS)’. Google offer GoogleApps, Microsoft offer online versions of Office, and so on. The second is ‘Platform as a Service (PaaS)’, typically offered through an API which integrates within one’s application. The third is ‘Infrastructure as a Service (IaaS)’, offering servers, databases, firewalls and routers as virtual appliances. The latter is the main topic of this write-up.

Benefits

Minimise Investment Risk – If an application’s success is short-lived or does not deliver, you won’t have invested in extra capacity needlessly. When pushing an application to the cloud, scalability and the risk of purchasing too much or too little infrastructure becomes the cloud provider’s issue. The cloud provider typically has such a massive amount of clients that it can absorb the growth and workload spikes of individual customers, reducing the financial risk they face.

Benefits which can also be reaped by using in-house virtualisation (e.g. VMware).

Speed to Deployment – Increasing infrastructure capacity can potentially become a matter of minutes.

Availability – As a side effect of adapting one’s application to the cloud, availability will improve and many  infrastructure SPoFs will be eliminated. Servers can fail, but the service will remain available.

Reliability – Applications are now built where entire servers can fail and not cause disruption, since server-specific data is irrelevant. This means that there will always be an active server ready to reply to a user’s request.

Flexibility and Agility – Cloud applications enables going to market very quickly.

Applicability

The classic example of parallelization with load balancing is a number of stateless Web servers all accessing the same data, where the incoming workload is distributed across the pool of servers. Cloud computing takes the concept to the next level.

Applicability to Databases – Database management systems have adapted to run in cloud environments by horizontally scaling database server and partitioning tables across them. This technique, known as sharding, allows multiple instances of database software to scale performance in a cloud environment by partitioning tables across multiple database instances, supporting large databases through horizontal scaling.

Applicability to CPU-intensive applications – Rather than creating a new thread for each task (example frame rendering), cloud optimized applications create a new virtual machines to render each frame, increasing performance through horizontal scaling

References:

http://www.sun.com/featured-articles/CloudComputing.pdf