Cluster Size Recommendations running 4.0 on Compute Engine

40-beta

#1

I’m running on Google Container Engine (GKE) based on the tutorial at https://github.com/couchbase/kubernetes. I’m looking for sizing suggestions. Since GKE is ultimately backed by GCE, any GCE guidance should apply

I’m trying to figure out the instance sizes to run and how many nodes to put in my cluster. I’m using it as a caching layer, so currently my reads and writes are about equal. I’m migrating from Google Cloud Datastore. This is what my workload has been averaging this month so far:

  • Avg 150 Write Ops / sec
  • Avg 150 Read Ops / sec
  • Peak 750 Write Ops / sec
  • Peak 750 Read Ops / sec
  • Avg 4kb record size
  • Peak 3 MB record size (rare)

I also expect to have 50-100GB persistant disk storage. What would be a good for the number of nodes and the machine type on GCE?

Thanks,
Derek


#2

Could you also tell us how many replicas you plan to use (your HA requirement) and if you are planning on running intensive queries and number of views and indexes you plan to add?


#3

I’m new to Couchbase, so I don’t know for sure. I’m running a multi-tenant SaaS system with namespacing in my current setup. I don’t know what the best way to namespace in CB will be when I’m not supposed to have over 10 buckets.

I think 1 replica with mostly key based access. Probably just one view that does some simple aggregation across time periods. I’m also wanting to run the sync gateway for mobile devices. Each user will have custom access to some subsection of the namespaced data.