Top three myths around the cloud
Until an application is rewritten, it is constrained as to what ‘cloud’ offering environment it can execute in. The fundamental commercial ‘cloud’ offering today is an IaaS (Infrastructure as a Service – see next myth) based upon virtualisation technology – making one server behave like multiple servers and allowing multiple applications that need dedicated servers to actually share a server. Many applications can be moved into this type of environment with the following critical caveats:
• Many applications require very specific types of I/O capabilities that may or may not be able to be virtualised. If not, then that application cannot execute in ‘the cloud’.
• Many applications require very specific hardware (i.e., NAS versus SAN). If so, then they can only run in virtualised environments that have that hardware and perhaps only if access to that hardware bypasses the virtualisation capability (effectively neutralizing the potential leverage and sharing of that hardware).
• Any application with real-time or near real-time performance requirements (i.e., high volume transaction processing) will likely fail in most existing cloud technologies.
Myth #3: Cloud offerings are highly segmented
There are really only three types of ‘cloud’ offerings: IaaS (Infrastructure as a Service), PaaS (Platform as a Service) and SaaS (Software as a Service). All the other ‘as a Service’ offerings you read about – such as Storage as a Service, Disaster Recovery as a Service, etc. – are specific versions of one of these three.
IaaS is what the name implies: a virtual infrastructure that stands in for (at a theoretically better price, performance, quality, and functionality point) a real infrastructure. The key point is the mapping to the real infrastructure. The virtual infrastructure is constrained to the capabilities of the real infrastructure it is ‘simulating’. If the IaaS does not support SAN, then the application cannot require a SAN.
IaaS is generally not micro scalable – that is, the IaaS is generally consumed in terms of its real counterparts (e.g., number of processor cores, GB of memory). Amazon EC3 is the best example of this – one selects from a menu of servers, system software and storage. This is effectively a SKU (Stock Keeping Unit) model. Other vendors offer similar SKUs with the added ability to build custom virtual infrastructures (the industry often uses the term ‘private cloud’), but not at the SKU costs. This is the model of the current Dell Services cloud – very specific configurations of hardware and software for specific prices. Also, today Dell can create a custom IaaS, but at a custom cost.
PaaS providers (Microsoft Azure, Salesforce.com, Google, and Joyent) basically provide a one-size-fits-all offering (with volume discounts and other pricing options). They do this because, in general, they make the hardware and system software (no choice in operating system, no choice in database system, no choice in transaction processing engine, etc.) totally invisible and inaccessible to the customer – applications have to be developed in and execute in their architecture and system software. Customers are allowed to develop in the PaaS environment (although the evolution is for third party vendors to create SaaS based upon the vendor’s PaaS). In effect, there is one basic SKU with some add-ons such as Disaster Recovery or higher Service Level Agreements (SLAs).
SaaS providers are yet one more step up in abstraction of the underlying technologies, in that they offer turnkey applications that generally run on someone else’s IaaS or PaaS, Salesforce.com and Google Apps being the exceptions. Customers have no access to the application code, though sometimes there are tailoring and reporting capabilities through some type of configuration management.
