mardi 22 septembre 2015

Spark in the Cloud - Usage only Pricing

I've been evaluating the costs associated with running Spark in the cloud.

Firstly I checked out Mesosphere. Using their template the estimated monthly cost was over 1k. This is mostly due to Amazon's pricing model, where it does not matter how much CPU you use but rather how long the instance is active.

Which means that if you don't want to be charged for unused CPU time, the cluster must be shutdown (and started again later).

I decided to check Google Cloud for a better deal. It seems that they charge by the minute rather than the hour, which is nice, but I believe they have the same model as Amazon when it comes to instance usage. ie. If you don't want to be charged shut it down.

I did see some info on how easy it should be to do the shutdown-restart process.

I understand they are charging for guaranteed CPU time rather than used CPU time, but there should be an alternative for users like me.

Is there an option that charges only for used CPU time running Spark workloads?

  • Can be any cloud provider.
  • Can use Mesos or similar tool, or none at all.

Aucun commentaire:

Enregistrer un commentaire