The "zero-copy" architecture is a clever marketing term for a real technical challenge. They claim to process identity graphs without moving raw PII data out of the customer's cloud environment.
Key technical claims:
* Compute is sent to the data (via containers) in your own VPC.
* Only derivative data (e.g., graph IDs, segments) are exported to their platform for orchestration.
* Avoids cloud egress costs and simplifies compliance scope.
I'm skeptical. The cost and complexity just moves.
* You now run their containerized workloads. What's the resource footprint? Have to right-size those pods.
* Network traffic shifts to cross-AZ, which still has costs.
* You pay for the compute and storage they use inside your account.
Need to see a concrete cost breakdown versus a traditional CDP. Has anyone run a Kubecost report or similar on a proof-of-concept deployment for this model? The real comparison is total cost of ownership, not just data transfer fees.
Example resource request from one vendor's sidecar collector:
```yaml
resources:
requests:
memory: "256Mi"
cpu: "250m"
limits:
memory: "1Gi"
cpu: "1000m"
```
Multiply that across nodes and clusters. It adds up.
null
You're right, the cost shifts. I trialed their setup and the sidecar collectors ate more compute than expected during peak event ingestion. Our Kubecost report showed it was cheaper than our old CDP's egress fees, but the hidden cost was our engineering team's time tuning those pods for a week. Not free, just different.
Still, from a compliance angle, keeping raw PII internal was a huge win for our legal team. That's the real trade-off, I think.
Another tool to try!