The stack
Every application sits on the same nine layers
From the application down to the network cables: on your own premises you buy, run and patch all nine layers.
Before the names, look at what is underneath any application. At the top is the application itself, the program people use. Under it is its data. The application needs a runtime, the engine for the language it was written in. Around that sits middleware, such as web servers and queues. All of it runs on an operating system. The operating system runs inside a virtual machine. That runs on a physical server, with storage, joined by a network. In your own building, every one of these nine layers is yours: to buy, to install, to patch, and to fix at night. Only the top one or two are the reason your company exists.
Service models
A service model is a line through the stack: above it is yours
IaaS, PaaS and SaaS differ only in how many layers the provider runs for you.
Cloud service models are one idea drawn four times. On premises, all nine layers are purple: yours. With infrastructure as a service, the provider takes the bottom four, the hardware and the virtualisation. You get a virtual machine and run everything from the operating system up. With platform as a service, the provider also takes the operating system, the middleware and the runtime. You bring only your application and its data. With software as a service, the provider runs the whole stack, including the application. You just use it. The names are less important than the line. Whenever you meet a new service, ask where its line is.
IaaS: infrastructure as a service
Rent the machine, run everything on it
Infrastructure as a service gives you virtual machines, disks and networks. The operating system and everything above it are your work.
Infrastructure as a service is the closest to owning a server. You ask for a virtual machine, and about a minute later you have one, with a disk and a network address. From there it is all yours. You install the operating system updates. You install the runtime and your application. You decide how it scales, and you set up the backups. On Amazon this is EC2, on Azure it is Virtual Machines, on Google Cloud it is Compute Engine. Reach for it when you are moving an existing system as it is, or when the software insists on a particular setup.
PaaS: platform as a service
Bring the code, the platform runs it
Platform as a service deploys, patches and scales for you. You own the code and the data.
Platform as a service moves the line up to just under your code. You push your application, and the platform puts it on servers, keeps the operating system patched, and adds capacity when traffic grows. You never log in to a machine. What remains yours is the code, the data, and a handful of settings. Elastic Beanstalk, App Service and App Engine are the application platforms. Managed databases belong here as well: you bring the tables, and the provider runs the database engine. Reach for it for anything new, when the team would rather build features than look after machines.
SaaS: software as a service
Finished software, you just use it
Software as a service is a complete application run by its vendor. Your data and your accounts are still yours.
Software as a service puts the line above the application. There is nothing to build or deploy. You open a browser and the software is there, usually for a price per user per month. Three things still belong to you: the data you put in, who has an account, and the settings, such as who may share a file outside the company. Microsoft 365, Google Workspace and Salesforce are the familiar examples. Reach for it whenever the need is the same in every company.
SaaS: software as a service · Going deeper
Every “as a service” name puts the line somewhere
CaaS, FaaS, DBaaS and DaaS are finer marks on the same scale. Serverless describes services with no servers to manage and no charge when idle.
Control and convenience
More done for you means less you can change
From IaaS to SaaS you trade control for convenience. Containers and functions are steps on the same line.
Put the models on one line and the trade is easy to see. On the left you control everything: the hardware, the operating system version, every setting. On the right the provider does almost all the work. As one wedge grows, the other shrinks. There are more than three stops. Between virtual machines and platforms sit containers, where you bring a packaged image. Between platforms and finished software sit functions, where you bring the code for a single task and nothing else. The useful habit is to start at the right and move left only when you find something you really need to control.
Shared responsibility
The provider secures the cloud; you secure what you put in it
The shared responsibility model: security of the cloud is the provider’s, security in the cloud is yours.
Handing over layers raises a question: who is responsible when something goes wrong? The answer has a name, the shared responsibility model. The provider is responsible for the cloud itself. That is the buildings, the power, the hardware, and the software that runs the services. Amazon calls this security of the cloud. You are responsible for what you put in it. That is your data, who can sign in and what they may do, how your network and firewalls are set, and whether your data is encrypted. Amazon calls this security in the cloud. The provider will run its half very well. It will not check your half for you.
Shared responsibility by model
The line moves with the service model; three rows never do
From IaaS to SaaS the provider takes on more, but data, devices and accounts stay with you in every model.
How much is yours depends on the model, and this table is how Microsoft draws it. Start at the bottom. In any cloud, the physical hosts, network and building are the provider’s. The middle rows move. On infrastructure as a service the operating system, the network controls and the applications are yours. On a platform, the provider takes the operating system, and the rows above become shared. On finished software, the provider takes nearly all of them. Now the top. Your information and data, the devices your people use, and their accounts are yours in every column. Nothing you buy changes that.
Shared responsibility by model · Going deeper · 1 of 2
Who patches what, service by service
For a virtual machine you patch the operating system. For a managed database, storage or functions the provider does, and access is still yours.
Shared responsibility by model · Going deeper · 2 of 2
Inherited, shared and customer-specific controls
AWS’s three kinds of control: those you inherit, those both sides apply to their own layer, and those only you can decide.
What is always yours
Your data, your identities and your devices are always yours
Breaches in the cloud are almost always on the customer’s side: an open bucket, a weak sign-in, a leaked key.
So what does that leave you with in practice? Three things, always. Your data: what you store, who may read it, whether it is encrypted and backed up. Your identities: the accounts, how people sign in, and what each one is allowed to do. Your devices: the laptops and phones that connect to all of it. This is where things go wrong. The analyst firm Gartner predicted that through 2025, ninety-nine percent of cloud security failures would be the customer’s fault. Attackers do not need to break into a provider. They look for the storage left public, the password with no second factor, the key pasted into public code.
What is always yours · Going deeper
The open bucket: nothing was hacked, the door was open
A misconfiguration is a setting left unsafe. It is the most common cause of data exposure in the cloud, and it is the customer’s side of the model.
Compliance
A certified cloud certifies the floor, not what you build on it
Compliance is shared too: you inherit the provider’s audited controls and are audited yourself for your half.
The same split applies to rules and audits. Providers are audited constantly against standards such as ISO 27001, SOC 2 and the payment card standard, PCI DSS. Those certificates cover their half: the buildings, the hardware, the services. Your half is your application, your settings, and your people. No one audits that for you. What you do get is evidence. Each provider has a portal where you can download its audit reports and hand them to your own auditor. Amazon’s is called Artifact, Microsoft’s is the Service Trust Portal, and Google has the Compliance Reports Manager.
Choosing a service model
Take the most managed option that fits
SaaS for common needs, PaaS for your own code, IaaS when you must control the operating system. Most companies use all three.
To choose between the models, ask three questions in order. Is this something every company needs, like email or payroll? Then buy it as finished software. If not, is it your own code, and would a standard runtime do? Then use a platform, or functions, and never touch a server. If not, does it need its own operating system setup, because it is old, or licensed in a certain way? Then rent machines and run it yourself. The order matters. Every layer you keep is a layer you patch and answer for. Keep the ones that make your product different, and hand over the rest.
Lesson complete
You finished “Service models and shared responsibility”
Nicely done.