Adding a Node to ceph tanks performance

[maomaocake]

I had a 3 node cluster where the command

rados bench -p scbench 10 write --no-cleanup

gets the following results


after adding another node however the results seem to have tanked.




I can obviously see the cur MB/s had a sharp drop, however Im not sure what is the issue and not even sure how to start debugging. Any help would be appreciated. Thanks

 

[maomaocake]

Here are the results of iperf3 to the other nodes (Ceph has its own 10G network)

 

[Maximiliano]

Hello,

Could you measure the latency between nodes using `ping`?

 

[maomaocake]

Sure, here are the results: