Compute cluster

This article provides a guide on how to create a compute cluster in Upsolver.

Add a compute cluster

1. From the Clusters page, click New > Compute.

2. Name this cluster.

3. (Optional) Select your AWS region.

4. (Optional) Select your private VPC.

Check Elastic IPs to create Elastic IPs for these servers.

The server instances will use these Elastic IPs allowing you to open access for your servers in external resources.

5. Select your server compute units to define the compute units one server will use. Choosing a server with 'high-memory' indicates a server type which has less CPU units but has more RAM.

6. Select a range of processing units between 1 and 64.

7. (Optional) Select a range for additional processing units for replay from 0 up to 64.

When configured, replay tasks will run on a separate cluster up to this size, which is billed separately.

If active, it is recommended for this cluster to be at least as big as the maximum size of the cluster.

Note: Replay cluster instances operate independently and have different IP addresses than the main cluster. Usually, tasks that write to external resources like databases are executed on the main cluster to maintain IP whitelisting. If you run into access issues, consider either turning off the replay cluster or using a separate, dedicated cluster.

8. Choose a scaling strategy.

9. Check Allow Maintenance Access if you wish to allow Upsolver to access your instances over SSH for maintenance purposes.

10. Click Create.

Scaling Strategy

When selecting scaling options, consider the following:

Scale Up

• Low Cost

  • looks exclusively at the work backlog, so it should reach 100% CPU and after a while scale up.

• Low Latency

  • scales up if the average CPU is over 80%.

• Consistent Low Latency

  • scales up if the average CPU is over 60%.

Scale Down

• Low Cost and Low Latency

  • check that the expected CPU after scaling down is below 70%.

    • This means that if there are 3 servers, they must be below 47% CPU.

      • $47 \cdot 3 = 141 \longrightarrow 141 / 2 = 70.5$

      • or 70.5% CPU for 2 servers

• Consistent Low Latency

  • applies the same checking, but the threshold is 50%.

  • This logic is to prevent scaling up and down in rapid succession.