In this post I will look at the three different service models for cloud computing as defined by NIST. More specifically I will look at the management and operations overhead for each one of the models and compare it to the traditional on-premise model.
Let's look at how things have been done in the past. Traditionally enterprises have been responsible for managing their own IT infrastructure as well as the software stack that runs their applications. For small companies that meant hiring polyglot employees with wide range of skills varying from low level networking to high level application support. For larger ones, that can afford more staff, it meant creating specialized teams responsible for only networking infrastructure, or only storage or servers and virtualization. However for lot of those enterprises the core business has never been managing IT infrastructure - the only thing they are interested in is to manage their line-of-business (LOB) apps.
Here are just some of the tasks IT teams enterprises have been required to do in the past:
- Build racks with servers and wire them into the network
- Build storage arrays and wire them into the network
- Configure routers
- Configure firewalls and DMZ zones
- Install operating system software on the servers
- Create virtual machines (if virtualization is utilized)
- Install operating system software on the virtual machines
- Install databases, set up replication and backups
- Install middleware used for hosting the application code
- Patch and update operating system software
- Patch and update databases
- Patch and update middleware
- Patch and update runtime
- Install application software
- Patch and update application software
Although long this is by no mean the complete list of tasks that IT personnel has been responsible for. From the list above only the last two (install application software and update application software) have been essential to the core business of the enterprise. In addition to the IT operational costs (OpEx) enterprises also incurred significant capital expenditures (CapEx) used to procure the necessary hardware.
More than a decade ago hosting providers recognized the need to help businesses with those tasks and allowed them to outsource the build-up of infrastructure, and concentrate on just managing their applications. Although hosting providers helped enterprises with OpEx and CapEx they still lacked some of the essential cloud characteristics like on-demand self-service, rapid elasticity and measured service as outlined in Essential Cloud Computing Characteristics.
Infrastructure-as-a-Service (IaaS) Model
IaaS model was the first model that complies with NIST cloud computing characteristics. In essence it offers cloud computing environment consisting of virtual machines. It offers self-service portal where you can on-demand start a virtual machine with preferred operating system, it is broadly accessible, elastic (you can easily start identical virtual machine or shut down exiting one), it uses pool of virtual machines that are collocated on common hardware, and at the end it measures your usage of those virtual machines.
If you look at the picture above you will see that the IaaS model provides automation in the lower layers (up to the virtualization layer) of the application stack. What that means is that tasks like starting the virtual machine, adding it to the network, configuring the routing and the firewalls and attaching storage to it is automatically done by the automation software. The vendor that provides the service is also responsible for managing any hardware failures and service the underlying hardware.
As you have already noticed the IaaS model provides cloud services up to the virtualization layer of the application stack. However as a consumer of the IaaS service you are still responsible for managing the virtual machine. Hence you are still responsible for patching and updating the operating system on the VM, installing and maintaining any databases or middleware that your application uses in addition to maintaining your actual application.
IaaS is very similar to the traditional hosting model with the added benefits of self-service, elasticity and metering.
Platform-as-a-Service (PaaS) Model
With PaaS you have much less things to worry about. As you can see from the picture the whole stack needed for your application is managed by the vendor. Your only responsibilities are your application and the data your application uses. In addition to the tasks managed in IaaS case the vendor (or in the case of private PaaS the platform owner) is also responsible for patching and updating the operating system, installing and maintaining the middleware as well as the runtime your application uses.
One important thing that you need to be aware of when using PaaS is that the automatic updates the vendor does may sometimes have negative impact on your application. Why is that? Very often OS and middleware vendors do incompatible changes between versions of their software. If your application depends on any underlying OS and middleware functionality it may break between platform updates. And because you are not in control of those updates you may end up with your application being down.
The premise of PaaS though is not only to offer maintenance free application stack but also additional services that you can utilize in your application. Very often PaaS providers are exposing middleware and databases as services and abstract the connectivity to those through APIs in order to free up developers from the need to locate the actual systems. Additional services can be authentication and authorization, video encoding, location based services etc. Using the PaaS services will allow you to abstract your applications from the underlying stack and as long as the APIs are kept intact it will be protected from failures between platform updates.
Software-as-a-Service (SaaS) Model
SaaS is the model with the highest abstraction and offers the most maintenance free option. As a SaaS consumer you are just using the software offered by the vendor. As depicted on the picture the whole stack is maintained by the vendor. This includes also updates for the application as well as application data management. SaaS model is very similar to the off-the-shelf software model where you go and buy the CD, install the software and start using it.
Traditionally one of the hardest problems application developers had to deal with was the data migration between different versions. SaaS vendors are also responsible for migrating your data and keeping it consistent. Similar to the off-the-shelf software model you can rely that you can access and read your data once you upgrade to a new version.
The SaaS model is the most resource-efficient model because it utilizes application multi-tenancy. What this means is that the same application instance handles multiple user-organizations. This is good for both the vendor and the customer because better resource utilization brings the maintenance costs down and hence the price for the services down. On the other side though tenant data is comingled and there is the security risk of one tenant accidentally getting access to another tenant's data.
Although not exhaustive the cloud computing service models explanation above should be enough to kick-start your initial discussion about your cloud strategy.