Citus storage best practice

[Ali Taher]

Hello,

We are currently testing citus in our lab consisting of 2 servers with the followings specs each:

• 32 cores (2 threads each)
• 256 GB RAM
• 10 SSD disks, 1.9 TB each
• 10 Gbit Network

we are using VMware ESXi on each of the servers where 7 VMs are created:

First server:

• 1 coordinator node : 32 vcpus , 64 GB RAM (VM created on first server, we are planning to add more workers nodes on this server later)

Second Server:

• 6 worker nodes : 4 vcpu , 32 GB RAM each (VMs are created on second server)

I'm not able to find any recommendation in the citus docs regarding the best approach for storage.

Here are the approaches I can think of:

1. Create one RAID 10 datastores and place all the worker nodes in it
2. Create multiple RAID 0 datastores and place one worker node per datastore
3. Don't use RAID at all and use 1 disk per worker node instead

Point 1 is more applicable for redundancy purpose, but does replication in citus solve the redundancy problem and then we can go with RAID 0 or no RAID at all for better throughput ?