I looked around but I didn’t fine any information…
What are the Minimum HW Requirements (number of machines, CPUs, RAM) for a k8ssandra and JanusGraph local cluster with Ubuntu 22.04 as common O.S.?
I looked around but I didn’t fine any information…
What are the Minimum HW Requirements (number of machines, CPUs, RAM) for a k8ssandra and JanusGraph local cluster with Ubuntu 22.04 as common O.S.?
I don’t know about JanusGraph, but a minimum of 3 Cassandra nodes is recommended for any production cluster to provide proper levels of availability and consistency.
HW requirements for Cassandra highly depend on the load you plan on putting on the cluster.
My personal recommendation for Cassandra nodes in production is 8 to 16 cores and 32GB to 128GB RAM, with up to 4TB disks.
I’m not sure what you mean by “local” when you say “JanusGraph local cluster”. Is it running on an actual Kubernetes cluster with worker nodes or do you want to run this with Kind/Minikube on a single machine?
Hi Alexander,
sorry for my misleading and not correct word “local”.
I meant an actual Kubernetes cluster with worker nodes on a cloud setting different from Azure, Google Cloud, Amazon
According to his: Hardware Choices | Apache Cassandra Documentation
minimun requirements for a cassandra node are:
at least 8 cores, at least 32 GB of RAM, more CPUs are better, Cassandra’s heap: at least 2 GB and no more than 50% of the total RAM
Why 2 machines with each 10 vCPU cores, 60 GB RAM, and 2 TB SSD are not enough ?
Why 2 machines with each 10 vCPU cores, 60 GB RAM, and 2 TB SSD are not enough ?
Because with 2 nodes and RF=2 (which is what you probably want to use) then you cannot use strong consistency (LOCAL_QUORUM
). If you do this, when a node goes down (due to an outage or a simple restart for maintenance) then all queries will fail.
If you choose to use LOCAL_ONE
as consistency level instead, you’ll have availability (one node can go down), but you don’t have guarantees that you can read the data you just wrote straight away (stale reads).
With 3 nodes and RF=3, you can lose one node and still achieve quorum, while ensuring you won’t have stale reads.
IMO, anything below that will remove a lot of the interest Cassandra has, probably meaning it’s not the right database for your use case.
Thank you very much Alexander
Very kind