How Alibaba Cloud’s Function Compute Self‑Select Specs Slash Serverless Costs

At the 2022 Hangzhou Cloud Expo, Alibaba Cloud announced a comprehensive price reduction for its Function Compute (FC) service, lowering vCPU unit prices by 11% and other billing items by up to 37.5%. The price cut is driven by a new technical capability: the function specification self‑selection feature.

Why the feature matters

Previously, FC allocated resources proportionally to memory size, forcing users to over‑provision CPU, memory, or disk to meet the most demanding dimension of a workload. This caused unnecessary cost and limited flexibility, especially for workloads with diverse resource profiles such as CPU‑intensive or I/O‑intensive tasks.

Self‑select specification

The new feature removes the fixed CPU‑memory ratio, allowing users to choose CPU, memory, GPU, and disk independently with fine‑grained steps. This enables precise matching of resource needs, higher utilization, and lower expenses.

Cost‑saving example

A function that consistently uses less than 1.5 vCPU and 6 GB memory previously required an 8 GB memory (4 vCPU) configuration, costing 0.0006096 CNY per second. After adjusting to 1.5 vCPU and 6 GB memory, the cost drops to 0.0002662 CNY per second—about 44% of the original expense.

Before: 4 vCPU * 0.000127 CNY/(vCPU·s) + 8 GB * 0.0000127 CNY/(GB·s) = 0.0006096 CNY/s

After: 1.5 vCPU * 0.000127 CNY/(vCPU·s) + 6 GB * 0.0000127 CNY/(GB·s) = 0.0002662 CNY/s

Tip: Monitor instance metrics; if vCPU or memory is under‑utilized, adjust the function specs to reduce cost.

Fine‑grained GPU allocation

For inference workloads needing only 1.8 GB GPU memory, traditional cloud servers would allocate an 8 GB GPU, resulting in <25% utilization. With Function Compute, users can allocate as little as 2 GB GPU memory, achieving near‑full utilization and elastic scaling without extra orchestration.

Specification limits

vCPU: 0.05 – 16 cores, in 0.05‑core increments

Memory: 128 MB – 32 GB, in 64 MB increments

GPU: 2 GB – 16 GB, in 1 GB increments

Disk: 512 MB (free) or 10 GB

These limits are configurable via the Function Compute console or programmatically with Serverless Devs by setting memorySize, cpu, and diskSize (memory and disk in MB, CPU in cores).

Underlying technology

Previously, FC used Docker containers on dedicated virtual machines per tenant, leading to resource waste because even a tiny 128 MB function required a 2‑core, 4 GB VM. The new architecture runs functions on elastic bare‑metal servers using secure containers, allowing >2000 function instances per server and dramatically higher density.

To support self‑select specs without degrading scheduling efficiency, FC builds a resource profile for each function, tracking its CPU, memory, GPU, and disk usage. During scheduling, the system selects machines where the new instance complements existing workloads, balancing resource consumption across the pool.

Combined, high‑density deployment and intelligent scheduling enable the flexible spec feature while preserving platform utilization and cost efficiency.