> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bidsmith.pro/llms.txt
> Use this file to discover all available pages before exploring further.

# Scaling

> Scale ARIS workers and registry components for higher throughput and reliability.

## Scaling goals

* Increase throughput without unstable tail latency
* Maintain request success rate during demand spikes
* Keep credit settlement and node discovery consistent

## Worker scaling strategy

<Steps>
  <Step title="Scale out first">
    Add more workers before increasing per-node concurrency.
  </Step>

  <Step title="Tune concurrency">
    Raise `max_concurrency` incrementally while tracking P95 latency.
  </Step>

  <Step title="Pin workloads">
    Route model families to dedicated node pools when possible.
  </Step>
</Steps>

## Registry scaling strategy

| Component | Recommendation                                                   |
| --------- | ---------------------------------------------------------------- |
| API layer | Run multiple stateless replicas behind a load balancer.          |
| Database  | Use managed PostgreSQL with automated backups and read replicas. |
| Cache     | Add Redis for node discovery and session lookup hot paths.       |
| Queue     | Use durable queues for asynchronous settlement and retries.      |

## Autoscaling signals

Use a combination of:

* queue depth
* in-flight request count
* P95 latency
* GPU utilization

<Accordion title="Avoid common scaling mistakes">
  <Warning>
    Scaling only by CPU can under-provision GPU-bound workloads.
  </Warning>

  <Warning>
    Large concurrency jumps can increase timeout rates and reduce total throughput.
  </Warning>

  <Warning>
    Mixing dissimilar models in one pool can create noisy-neighbor effects.
  </Warning>
</Accordion>