This article compares AWS Lambda and virtual machines, discussing when to use each and digging into pricing.

A virtual machine isn’t the only way to get computing power on AWS, and it isn’t always the most cost-effective. Sometimes you just need to set up a service that will perform a task on demand, and you don’t care about the file system or runtime environment. For cases like these. AWS Lambda may well be the better choice.

Amazon makes serious use of Lambda for internal purposes. It’s the preferred way to create “skills,” extended capabilities for its Alexa voice assistant. The range of potential uses is huge.

AWS Lambda and virtual machines both exist on a spectrum of abstraction wherein you take on less and less of the responsibility for managing and patching the thing running your code. For this reason, Lambda is usually the better bet when your use case is a good fit.

What is AWS Lambda

Note: For a full analysis breaking down what AWS Lambda is with pricing examples, see our earlier post, “What is AWS Lambda – and Why You’re About to Become a Huge Fan“.

Lambda is a “serverless” service. It runs on a server, of course, like anything else on AWS. The “serverless” part means that you don’t see the server and don’t need to manage it. What you see are functions that will run when invoked.

You pay only per invocation. If there are no calls to the service for a day or a week, you pay nothing. There’s a generous zero-cost tier. How much each invocation costs depends on the amount of computing time and memory it uses.

The service scales automatically. If you make a burst of calls, each one runs separately from the others. Lambda is stateless; it doesn’t remember anything from one invocation to the next. It can call stateful services if necessary, such as Amazon S3 for storing and retrieving data. These services carry their own costs as usual.

Lambda supports programming in Node.js, Java, Go, C#, and Python.

Comparison with EC2 instances

When you need access to a virtual machine, Amazon EC2 offers several ways to obtain one. It has three ways to set up an instance which is a VM.

On-demand instances charge per second or per hour of usage, and there’s no cost when they’re inactive. The difference from Lambda is that the instance is a full computing environment. An application running on it can read and write local files, invoke services on the same machine, and maintain the state of a process. It has an IP address and can make network services available.

Reserved instances belong to the customer for a period of time, and billing is for the usage period. They’re suitable for running ongoing processes or handling nearly continuous workloads.

Spot instances are discounted services which run when there is spare capacity available. They can be interrupted if AWS needs the capacity and will pick up later from where they left off. This approach has something in common with Lambda, in that it’s used intermittently and charges only for usage, but it’s still a full VM, with all the abilities that imply. Unlike Lambda, it’s not suitable for anything that needs real-time attention; it could be minutes or longer before a spot instance can run.

Abstract 3D illustration. Mathematical symbol lambda; as a chip with electronic tracks in a puzzle of glass.

Use cases for Lambda

Making the right choice between AWS Lambda and virtual machines means considering your needs and making sure the use case matches the approach.

The best uses for Lambda are ones where you need “black box” functionality. You can read and write a database or invoke a remote service, but you don’t need any persistent local state for the operation. Parameters can provide a state for each invocation. Cases which this functionality could be good for include:

  • Complex numeric calculations, such as statistical analysis or multidimensional transformations
  • Heavy-duty encryption and decryption
  • Conversion of a file from one format to another
  • Generating thumbnail images
  • Performing bulk transformations on data
  • Generating analytics

Invoking a Lambda service is called “triggering.” This can mean calling a function directly, setting up an event which makes it run, or running on a schedule. With the Amazon API Gateway, it’s even possible to respond to HTTP requests.

AWS Step Functions, which are part of the AWS Serverless Platform, enhance what Lambda can do. They let a developer define an application as a series of steps, each of which can trigger a Lambda function. Step Functions implement a state machine, providing a way to get around Lambda’s statelessness. Applications can handle errors and perform retries. It’s not the full capability of a programming language, but this approach is suitable for many kinds of workflow automation.

AWS Lambda and Virtual Machines | Comparing costs

Check out Amazon’s pricing calculators for full details on Lambda pricing and EC2 pricing.

Like other factors when comparing AWS Lambda and Virtual Machines, Lambda wins out on cost if your use case supports using it.

Lambda wins on cost when it’s employed for a suitable use case and when the amount of usage is relatively low. “Relatively low” leaves a lot of headroom. The first million requests per month, up to 400,000 GB-seconds, are free. Customers that don’t need more than that can use the free tier with no expiration date. If they use more, the cost at Amazon’s standard rates is $0.0000002 per request — that’s just 20 micro cents! — plus $0.00001667 per GB-second.

The lowest on-demand price for an EC2 instance is $0.0058 per hour. By simple division, neglecting the GB-second cost, a Lambda service can be triggered up to 29,000 times per hour and be more cost-effective.

Many factors come into play, of course. If each request involves a lot of processing, the costs will go up. A compute-heavy service on EC2 could require a more powerful instance, so the cost will be higher either way.

Some needs aren’t suitable for a Lambda environment. A business that needs detailed control over the runtime system will want to stay with a VM. Some cases can be managed with Lambda but will require external services at additional cost. When using a virtual machine, everything might be doable without paying for other AWS services.

The benefits of simplicity

When it comes to AWS Lambda and virtual machines, it comes down to using the simpler method as long as it meets your needs. If a serverless service is all that’s needed, then the simplicity of managing it offers many benefits beyond the monthly bill. There’s nothing to patch except your own code, and the automatic scaling feature means you don’t have to worry about whether you have enough processing power. It isn’t necessary to set up and maintain an SSL certificate. That frees up IT people to focus their attention elsewhere.

With the Lambda service, Amazon takes care of all security issues except for the customer’s own code. This can mean a safer environment with very little effort. It’s necessary to limit access to authorized users and to protect those accounts, but the larger runtime environment is invisible to the customer. Amazon puts serious effort into defending its servers, making sure all vulnerabilities are promptly fixed.

With low cost, simple operation, and built-in scalability, Lambda is an effective way to host many kinds of services on AWS.