There is a lot of buzz and marketing around the term “serverless”. This article aims to dismantle what’s serverless and what not - with practical examples.
Definition
So first, we need a definition of serverless. A good start might be Wikipedia:
Serverless computing is a cloud computing execution model in which the cloud provider allocates machine resources on demand, taking care of the servers on behalf of their customers. […] However, developers of serverless applications are not concerned with capacity planning, configuration, management, maintenance, fault tolerance, or scaling of containers, VMs, or physical servers. […] When an app is not in use, there are no computing resources allocated to the app. Pricing is based on the actual amount of resources consumed by an application.[1] It can be a form of utility computing.
This definition is very tightly coupled to compute, while a backend application usually consists of many more components like storage, queues, databases, event busses, load balancers, etc.
Paul Johnston, one of the fathers of ServerlessDays, has a more broad definition, which doesn’t only apply to compute, but to software solutions in general.
A Serverless solution is one that costs you nothing to run if nobody is using it (excluding data storage).
The inherent essence of this definition is that every building block of a serverless solution MUST scale to zero (or isn’t billed) if not used. Data storage is an exception since it’s usually not wanted that data are wiped automatically if not used, unless it’s a cache.
So let’s have a look at use cases or Public cloud services and see if they meet the serverless definition above:
| Product / use case | is it serverless? | why? |
|---|---|---|
| Amazon Elastic Kubernetes Service (EKS) | ❌ | $40 base fee / month |
| Azure AKS | 🤔 | No base fee. But also apparently useless without worker nodes |
| Google Kubernetes Engine (GKE) | ❌ | $74 base fee / month |
| Knative | 🤔 | Depends:
|
| Serverless Aurora v1 | âś… | Can scale to zero |
| Serverless Aurora v2 | ❌ | Despite it’s name, it’s not serverless, since it cannot scale to zero and costs $43/month for idle. |
| Kinesis Data Streams | ❌ | Minimum 1 shard, $29/month |
| Kinesis Data Streams On Demand | âś… | On-demand, scales to zero |
| AWS, SQS, SNS, EventBridge | âś… | On-demand, pay-per-use |
| DynamoDB Standard | ❌ | Pricing based on provisioned capacity units |
| DynamoDB on demand | âś… | Per request pricing |
| AWS Elastic Beanstalk | ❌ | The ELB/ALB/NLB load balancer has a base fee |
| AWS App Runner | âś… | App Runner scales to zero |
| AWS ECS+Fargate | âś… | While Fargate can scale to zero, a E/A/NLB load balancer is most likely used. |
| Google App Engine Standard | âś… | scales to 0 |
| Google App Engine Flexible | ❌ | Despite its name, it’s not that flexible and needs VMs |
| Akamai Edge Functions | 🤔 | Need to contact sales |
I will add more examples as they cross my attention.
Ok, at this point you might be like “So what?”. In another article, we will have a look at why scaling to zero is actually a very important and good architectural characteristic.
Like what you read?
You can hire me or make a donation via PayPal!