Notion: API caching test methodology

index.md

I did a series of local tests to determine how quickly Notion would load if all API responses were available near-immediately. Full article → Case study: Analyzing Notion app performance

Methodology

Preparation

1. Use Charles as a local HTTP/2 proxy
2. Intercept and save API responses for the following requests:
   • POST /api/v3/loadPageChunk
   • POST /api/v3/getUserAnalyticsSettings
   • POST /api/v3/getClientExperiments
   • POST /api/v3/getPublicPageData
   • POST /api/v3/syncRecordValues
3. Using the “Map local” feature, make above requests serve saved responses from the local FS
4. Confirm that with “Map local” enabled, response times go down from 200+ ms to 20+ ms
5. Go to Chrome DevTools → Audits → Settings and uncheck the “Simulated throttling” flag

Running tests

6. With “Map local” disabled (but with Charles still running), run 3 performance audits
7. With “Map local” enabled, run 3 other performance audits

Extracting results

8. For each performance audit:
   • click “View trace”
   • go to Frames
   • find the first frame which has the page text
   • and record the timestamp of that frame

Results

Raw performance traces.

	Without “Map local”, ms	With “Map local”, ms
Audit 1	20887	18570
Audit 2	21472	19355
Audit 3	20685	18817
Average	21015	18914

Improvement: 9.9%