back_office_ops · workflow
Cannes: How Dropbox ML saves $1.7M a year on document preview pre-warming
Dropbox's Riviera system pre-warmed previews for all eligible files, but a significant portion of that pre-generated content was never viewed, generating considerable and unnecessary CPU and storage costs at petabyte scale.
How it works
Common implementation structure
How this type of workflow is generally built, generalized across documented cases — not tied to any one vendor's stack. Click any stage to read what happens there. Specific products that implement these stages appear in “Tools commonly seen” below.
Stage 1 · File enters pre-warm path
Riviera collects all file IDs eligible for pre-warming and sends a prediction request with the file ID and file type.
Tools used
RivieraSuggest BackendUser Profile ServicePredict ServiceStormcrowGrafanaSupersetRocksDBHive
Outcome
Cannes replaced an estimated $1.7 million in annual pre-warm costs with $9,000 in ML infrastructure per year after being deployed to almost all Dropbox traffic, with no observed degradation to preview latency.
Results
Volume$9,000
Cost replaced$1.7 million
Source
https://dropbox.tech/machine-learning/cannes--how-ml-saves-us--1-7m-a-year-on-document-previews
Grounding & classification
Source type: technical build writeup
24 fields verified against source quotes, 1 dropped as unverifiable.
predictive analyticsbuilder submittedfailure mode describedmetric backednamed customerproduction runtime claimedtools describedworkflow describedsoftwarecost reductiontechnical build writeupback office ops