42: Serverless is the answer to everything — Part 1

Co-Author: Emily Jiang

Serverless means you don’t need a server to run your applications, right? Recently I collaborated with the IBM Liberty Cloud Native Architect and Chief Advocate, Emily Jiang, to examine if serverless is indeed free computing.

What is Serverless?

Serverless has become a buzzword nowadays. You might be wondering what Serverless is exactly. Do we need to run all our cloud native applications as serverless? This articlee will help you with these questions.

Serverless is a short name for Serverless Computing, which is an execution model in which the cloud provider allocates resources on demand. Serverless enables developers to concentrate on their own applications without the need of worrying about servers. It does not mean there is no server running in the background. On a contrary, normally Serverless architecture contains servers. There are pros and cons for the Serverless Architecture.

Pros

• Cost effective for public cloud — Serverless architecture enables applications to scale to zero and it only uses resources when there is workload.
• No need to manage server — Serverless architecture hides the server and abstract away the maintenance of the servers.

Cons

• Not managing your own server limits your flexibility — Serverless abstract away the maintenance of servers, which means that you are limited to the services the cloud provider has.
• Vendor lock-in — Since your application chooses the services provided by a particular Cloud Provider, your Serverless deployment might not work on another Cloud Provider.

Serverless normally execute code in the form of a function. Therefore, serverless is sometimes referred to as “Functions as a Service” (FaaS). Below are the FaaS offering from major cloud provider:

• IBM Code Engine: IBM Cloud Function
• AWS: AWS Lambda
• Microsoft Azure: Azure Functions
• Google Cloud: Google Cloud Functions

You might be wondering whether Serverless is suitable for all kinds of applications or deployment environment.

What is Serverless for?

First, Serverless is a deployment architecture for public cloud infrastructure with the intend of cost saving. Serverless normally uses event to trigger your application to run. When there is no trigger, the application can scale to zero, which means no resource consumption. To respond to the event in a timely manner, the application must start fast. If it takes too long such as seconds or minutes to start up, it would have defeated the intend of Serverless. The Serverless model is like the Pay-as-you-go package for your mobile. When using Pay-as-you-go, you normally only send texts or making phone calls when there is absolute need. In this way, you will be able to keep the cost to the minimum.

If you have decided to go with serverless, the next question is that what kind of application architecture should be adopted.

• Microservice architecture — Since the application will run as a function, it works better to have a small dedicated application, such as microservice.
• Stateless — Since serverless will scale to zero when not invoked, there will be frequent start and stop. Every time on start, a brand-new container will be created, so none of the previous context will be kept.
• Fast — Serverless will need to start up and vamp-up fast so that some cost can be saved. If it takes too long to start up or shut down, serverless won’t be useful at all.

The cost of Serverless

There is a common misconception that “serverless is free”. The misconception is likely from the ideal behind “scale to zero”, i.e., if I can scale my services to zero, I am not paying anything when it is not being used. That is certainly true, provided you are not also the serverless infrastructure provider. To host a serverless infrastructure, one would have to run KNative and Kafka, both of which are not free — as in order to handle the serverless request, the serverless infrastructure must be running.

Another consideration is the purpose of a solution is to solve a business problem, not to “scale to zero”. We should always optimize first for when the service is being used, but not optimize for when it will not be used. “serverless” is not an excuse to write sub-optimized code or code that will not scale efficiently. For a service that will be used often or constantly, there is almost no benefits to serverless as the server will likely never “scale to zero”. Not to mention if you’re responsible for running the serverless infrastructure, you have to pay for the unnecessary infrastructure cost as well.

References

The original story was first published by the same authors in IBM TechXchange Community:

1. In the first article, we discussed what serverless is and whether the true cost of serverless.
2. In the second article, we examined a popular serverless implementation — Knative and discussed the use cases for “serverful” vs “serverless”

Hopefully, the articles will help you to decide if serverless is right for your applications and your specific use cases.