Please could you tell me how correctly calculate Arbitrator system requirements?
Is it possible to calculate the sizing of the Arbitrator (RAM, CPU, Disk resources) based on the members hosts resources which already in use in existing Percona cluster?
The goal to have enough resources for Arbitrator and no negative impact on the cluster perfomance at maximum load in future.
For example (if applicable)
- We have to have the same network throughput with minimum latency and RTT for both Arbitrator and Percona cluster host
- No need too much disk space as no mysql data writing actually happens by garbd and we don’t need SSD disk on Arbitrator in spite of SSD disks are used at Percona cluster hosts
- we need less CPU cores and RAM resources than on each Percona cluster members. Which % CPU Cores and RAM of existing Percona cluster node could be recommended for the Arbitrator?If there are any Arbitrator system requirements recommendations (best practice) please can you share the link?
For some reason i can’t find it but i believe the question should be very popular.
Arbiter needs to receive the same traffic as other Percona XtraDB Cluster nodes but it does not need to apply them. This means it has same network performance requirements as data nodes, rather small memory requirement and a lot less CPU requirements. I do not have exact number but I’d start with instances on smaller side and see if you see any bottleneck.
Peter, thanks!