Pause and recovery as one mechanism
A deliberate pause and a dropped connection leave the system in the same state. Building one mechanism instead of two removed a class of divergence between paths that are meant to behave identically.
Reliable transfer and migration systems
Resumable uploads, a faster migration path and a dedicated download service. All three sit where users notice reliability and speed straight away.
- Organisation
- Zoho Corporation
- Period
- 2018 – 2020
- Area
- Product engineering
Context
A notes product lives on moving user content in and out: uploading it, importing it from a competitor, downloading it back. Each is a place where a failure is immediately visible.
The problem
An interrupted upload that restarts from zero is a data-loss event as far as the user is concerned. A migration that runs through conventional per-item API calls is slow enough to deter people from switching at all. And download latency is felt on every single request.
Constraints
- Uploads had to survive interruption and user-initiated pauses without losing transferred data.
- Migration had to handle another product's export format and volume.
- Improvements had to work within the existing product.
My contribution
Shared ownership stated as such. Nothing here claims sole authorship of a platform.
- Built pause-and-resume support for file uploads, allowing interrupted transfers to continue without data loss.
- Built a migration path from Evernote to Zoho Notebook that avoided conventional API round-trips, reducing migration time.
- Designed a dedicated download-serving capability that optimised request handling and reduced download time by approximately 30%.
Approach
Make transfer state explicit and durable, so resuming means reading a record instead of guessing what arrived. It also means a deliberate pause and a dropped connection are the same operation.
Bypass the per-item API path for migration. The bottleneck was the number of round-trips, not the work inside each one, so tuning them individually would not have moved the result.
Separate download serving from general request handling, because a path with different characteristics is easier to optimise once it is no longer sharing one.
Decisions that mattered
The ones with a real cost on the other side.
Outcome
Transfers that survive interruption, a materially faster migration path from a competing product, and roughly 30% lower download time.
What generalises
- Resumability comes from state you deliberately recorded, not from inference.
- Optimise the dominant cost. The number of round-trips usually matters more than how fast each one is.