September 25, 2024•5 minute read
AWS Lambda vs. Google Cloud functions: a comprehensive comparison
Yiren Lu@YirenLu
Solutions Engineer
Both AWS Lambda and Google Cloud Functions offer serverless execution environments for building and connecting cloud services. They allow developers to write single-purpose functions that are triggered by events or HTTP requests, enabling rapid development and scalable applications.
This article provides an in-depth comparison of these two services, examining key factors such as features, performance, pricing, and integration capabilities.
Key features comparison
| Feature | AWS Lambda | Google Cloud Functions |
|---|---|---|
| Supported Languages | Node.js, Python, Java, Ruby, C#, Go, PowerShell | Node.js, Python, Go, Java, .NET, Ruby, PHP |
| Maximum Execution Time | 900 seconds (15 minutes) | 3,600 seconds (60 minutes) for 2nd gen |
| Pricing Model | Per-request and per 1ms of execution time | Per-request and per 100ms of execution time |
| Free Tier (monthly) | 1 million requests, 400,000 GB-seconds (memory) | 2 million requests, 360,000 GB-seconds (memory), 180,000 GHz-seconds (CPU) |
| Cold Start Mitigation | Provisioned Concurrency | Minimum Instances |
| VPC Support | Yes | Yes |
| Ecosystem Integration | Deep integration with AWS services | Deep integration with Google Cloud services |
| Event Triggers | Wide range of AWS service triggers | HTTP, Cloud Storage, Pub/Sub, Firestore, and 90+ via Eventarc |
| Monitoring | CloudWatch | Cloud Monitoring |
| Deployment Tools | AWS CLI, AWS SAM, CloudFormation | gcloud CLI, Cloud Console, Terraform |
| Maximum Memory | 10 GB | 32 GB |
| Maximum Concurrent Executions | 1,000 per region | 1,000 per function |
| GPU Support | No | Yes (in preview) |
Event-driven capabilities
Both platforms offer robust event-driven architectures:
AWS Lambda
Supports a wide range of trigger types, including:
- HTTP/HTTPS (via API Gateway)
- Amazon S3, DynamoDB, SQS, SNS
- AWS CloudWatch Events/EventBridge
- AWS CloudWatch Logs, CodeCommit, Kinesis
- AWS IoT, Alexa Skills Kit
Google Cloud Functions
Supports various trigger types:
- HTTP triggers
- Event triggers:
- Pub/Sub, Cloud Storage, Firestore
- Generalized Eventarc triggers (90+ event sources via Cloud Audit Logs)
Resource configuration and scaling
AWS Lambda
- Automatically allocates CPU power proportional to configured memory
Google Cloud Functions
Cold start mitigation
Both AWS Lambda and Google Cloud Functions offer strategies to mitigate cold starts:
AWS Lambda
- Provisioned Concurrency: Keeps a specified number of initialized instances ready to respond to invocations.
Google Cloud Functions
- Minimum Instances: Keeps a specified number of instances warm and ready to serve requests.
- CPU Allocation: Allocating more CPU can help reduce cold start times for compute-intensive functions.
Pricing structure
Both services use a pay-per-use model with some differences:
AWS Lambda
- Charges in 1ms increments after the first 100ms
- Pricing based on number of requests, memory allocated to the function, and execution duration
Google Cloud Functions
- Charges in 100ms increments
- Pricing based on invocations, CPU and memory allocation, and execution time
Pricing comparison
| Service | Free tier | Compute pricing | Request pricing | Additional charges |
|---|---|---|---|---|
| AWS Lambda | 1M free requests per month and 400,000 GB-seconds of compute time | $0.0000166667 per GB-second (includes CPU and memory) | $0.20 per 1M requests | $0.09 per GB outbound data transfer |
| Google Cloud Functions | 2M free requests, 180,000 vCPU-seconds, and 360,000 GB-seconds per month | $0.00002400 per vCPU-second (CPU) and $0.00000250 per GB-second (memory) | $0.40 per 1M requests | $0.12 per GB outbound data transfer |
Note:
- AWS Lambda pricing is based on the amount of memory you allocate to your function and the time it runs.
- Google Cloud Functions prices CPU and memory separately, allowing for more granular resource allocation.
- Prices shown are for Tier 1 regions in Google Cloud.
- Google Cloud Functions offers committed use discounts (CUD) for longer-term commitments, which can reduce costs further.
- For both services, actual costs may vary based on specific configuration and usage patterns.
GPU support
AWS Lambda
AWS Lambda does not offer native GPU support, but provides alternatives:
- Can trigger GPU-enabled EC2 instances or ECS tasks
- Integrates with AWS Batch for GPU-accelerated batch processing
- Works with Amazon SageMaker for ML inference on GPUs
Google Cloud Functions
- Offers GPU support in preview for Cloud Run functions (2nd gen). You can currently access only 1 Nvidia L4 GPU per Cloud Run instance.
- Can trigger Compute Engine instances with GPUs
- Works with Google Kubernetes Engine (GKE) for containerized GPU tasks
Please check the official pricing page for the most up-to-date and detailed information.
VPC support
VPC support is crucial for security and access to private resources:
AWS Lambda
- Native VPC integration
- Functions can run in private subnets
- Direct access to VPC resources without public exposure
- Security group association for traffic control
- Automatic ENI management
- Potential longer cold starts for VPC-connected functions
Google Cloud Functions
- Native VPC support (2nd gen only)
- Serverless VPC Access Connector feature for accessing VPC resources
- Primarily facilitates outbound connections to VPC
- Shared VPC support across projects