Serverless is cheaper for IoT-type use cases confirms 451 Research

Likely price war between the 'big four' will broaden the appeal of 'function-as-a-service'

A study by analyst 451 Research has found that serverless cloud services can save money, but only under certain conditions.

The last two years have witnessed a growning interest in serverless cloud, otherwise known as function-as-a-service (FaaS). Among other purported benefits, FaaS is promoted as offering cost savings, particularly for small, event-driven processes. This is because the customer is only charged for the time that the function is running, rather than for the entire uptime of a virtual machine.

Whether serverless actually turns out to be cheaper depends very much on what it's used for, the report finds. It is particulary suited to use cases such as IoT and AI, where a function executes in response to some trigger and then promptly shuts down again. The total running time - perhaps a few milliseconds per operation - is aggregated over the month and the customer is billed accordingly.

In such use cases savings can be made over the cost of simply provisioning a VM provided that the code is executed fewer than 5,000 times per month, 451 Research has found.

Further savings may be possible on developer time as there is less provisioning and admin required. 451 Research estimates that for a function with one-second duration operating for an aggregate total of three-quarters of a month, if the developer saves 10 minutes of time on provisioning duties in that month then savings will be made, all other factors being equal.

The flipside of this is that applications and functions that do not run for a short period of time following a trigger but which are more general in scope are likely to be better off hosted on a standard VM, PaaS or container if cost is the main issue.

There may be other considerations, too, such as the provision of memory. Most serverless providers require memory to be allocated to some code before it is run, which may result in over- or underprovisioning. Moreover, the ‘big four' serverless provers (Amazon Lambda, IBM Bluemix OpenWhisk, Google Cloud Functions and Microsoft Azure Functions) are all limited in the type of code they can run. Google supports only Node.js, while AWS Lambda is limited to Node.js, Python, Java and C#.

In terms of cost comparisons, each vendor has its 'sweet spot', but 451 Research notes that overall prices are roughly equivalent, suggesting that some serious competitive cost cutting is likely in the near future. It also foresees the emergence of serverless brokers, picking and choosing between the vendors to optimise price, automation or performance.

Serverless cloud will be an area of focus at the forthcoming Computing Cloud & Infrastructure Summit 2017 - registration is free for most delegates