Designing an API for Mobile
1 like•296 views
Victor Quinn, a lead Node.js engineer, discusses designing a mobile API with a focus on crucial factors such as speed, battery usage, response size, and minimizing the number of requests during a presentation at the DMV Meetup. Solutions include caching, batch requests, and utilizing a custom batch request library to enhance API performance and efficiency. The presentation emphasizes the importance of stateless interactions and optimizing for mobile network constraints.
Designing an API for Mobile
- 1. Victor Quinn @victorquinn Lead Node.js Engineer Designing an API for Mobile API Craft DMV Meetup, April 28 2014
- 2. Download Gives you real-time information on the people around you ! It’s free!! ! This will be pretty fast-paced More About SocialRadar
- 3. Special considerations for a Mobile API • Speed really matters • Battery usage - minimize • Maximizing network utilization • Response size matters • Minimize number of requests • Stateless
- 4. Speed: Problems • Only one thing on screen at a time, perception change • On desktop other things can distract • Best Case: ~400ms latency for round trip on cell network • We tested at SocialRadar using a basic ping no-op endpoint • Verizon LTE, iPhone 5S • Device doesn’t maintain active connection • Average case much worse, ~1000ms round trip
- 5. Ideal round trip iPhone 5S on Verizon LTE User's Phone API 200ms 200ms 200ms Total round trip time 800ms 200ms bit of hand waving, I’m not a network expert
- 6. Likely round trip iPhone 4S on Verizon 3G User's Phone API 500ms 500ms 200ms Total round trip time 1600ms 400ms
- 7. Speed: Solutions • Cache anything humanly possible • Respond as quickly as possible, deferring anything that doesn’t need to be done immediately • Queue most things • Make sure your API can handle high concurrency (we settled on Node.js to accomplish this, WhatsApp uses Erlang)
- 8. Minimize battery usage ! • Brief note about cell battery and network usage (more) • Response size matters • Minimize number of requests Maximize network utilization
- 9. Response Size: Problems • At SocialRadar, we tested responses of different sizes • Optimal was largest response that would fit into a single TCP packet on the cell network. • Around 128KB. Differs based on client device and network • Any longer would jump response time significantly
- 10. Response Size: Solutions • Paging sizes — largest that fit within TCP packet window • Contrast with Desktop, paging more dependent on server response time and UI
- 11. User's Phone API 200ms 200ms 200ms Total round trip time 800ms 200ms Page size of 20 users To retrieve 100 users, ~4s Paging Size: 20 (Ideal)
- 12. Paging Size: 50 (Ideal) User's Phone API 200ms 200ms 500ms Total round trip time 1100ms 200ms Page size of 50 users To retrieve 100 users, ~2.2s
- 13. Paging Size: 20 (Likely) User's Phone API 500ms 500ms 200ms Total round trip time 1600ms 400ms To retrieve 100 users, ~8s
- 14. Paging Size: 50 (Likely) User's Phone API 500ms 500ms 500ms Total round trip time 1900ms 400ms To retrieve 100 users, ~3.8s
- 15. Number of Requests: Problems • Traditionally parallelized, not (well) on mobile • Battery life is of great concern and every network request eats into that • Spinning up the cell modem, making request, waiting for it drains the battery • So if you make 5 requests instead of 1, draining battery, leaving user waiting
- 16. Number of Requests: Solutions • Paging size as mentioned above, try to maximize response that will fit in a packet to minimize the number of requests • So one request with 20KB much better than 5 with 4KB each • Batch multiple requests into a single one
- 17. Batch Request Library • Finding no great existing batch request library for Node.js, we rolled our own open source library • http://batch-request.socialradar.com • Allows you to send a single request that represents multiple • No need to compromise on RESTful principles and move to a SOA • Handles complex dependencies
- 18. Batch Request Library (2) 1. npm install batch-request 2. Add one line of code to your API: app.post('/batch', batch.validate, batch);! 3. Send batch request as single POST: { "myRequest1": { "method": "GET", "uri": "http://api.mysite.com/users/1/first_name" }, "myRequest2": { "method": "GET", "uri": "http://api.mysite.com/users/1/email" } }
- 19. Batch Request Library (3) 4. Receive single response: { "myRequest1": { “statusCode": 200, “body": “Victor”, “headers”: {…} }, "myRequest2": { “statusCode": 200, “body": “[email protected]”, “headers”: {…} } }
- 20. When you pull to refresh, Batch Request to: • Update user location with latest GPS reading • Retrieve current City/State • Retrieve most recent stat numbers • Get some user lists to cache them
- 21. Stateless • Cell networks are spotty • For performance, want load balance among multiple servers • Solution: Assign access token to device rather than relying on sessions for user access • We have an access token which is an encrypted string containing user_id and some other stuff
- 22. Victor Quinn @victorquinn Lead Node.js Engineer ! Download our App Questions?