Scaling

ScienceOps is built to scale 2 ways:

  • Vertically: Increasing the RAM and/or CPU of the cluster's servers.
  • Horizontally: Increasing the number of servers in the cluster.

Vertical

How to do it

You'll need to remove a worker from the cluster in order to scale it up. Begin by visiting the "Workers" tab on the /admin/servers page. Remove the worker from the set of available workers by clicking the "Remove Server" button. Once you've done that, you can resize the machine as you see fit. Once it's been resized, boot it back up. Visit the "Workers" tab on the /admin/servers page again and add the worker back to the cluster by using the "Add Server" button.

What to expect

The most common use-case for vertical scaling is adding RAM to the system. This is common when a model with a large memory footprint is deployed and you need to adjust the workers accordingly. ScienceOps will begin to use the resources for the resized worker immediately.

Horizontal

How to do it

Spin up a new server and install the scienceops-worker package and dependencies. Once setup, you can add the worker to your ScienceOps cluster via the admin console (see /admin/servers page, "Workers" tab). Enter the hostname/IP of the new worker and click the add worker button. You the worker's IP should appear in the list of workers.

What to expect

ScienceOps will begin to send new deployments to the new worker machine. It'll do so until the model containers are balanced out across all of the workers. ScienceOps does not rebalance the workers upon adding a new worker. To achieve this, simply restart ScienceOps and on boot, ScienceOps will balance out the models.

results matching ""

    No results matching ""