I want to understand the bandwidth requirement of our use case.
Currently we have got 4 nodes in our cluster with replication factor of 2. On any 2 nodes, I’ve verified that I’m getting around 5 Gbps TCP throughput. This I’ve verified using iperf.
Given this bandwidth, I want to know what throughput(read+write) can I achieve(theoretically). By read+write I mean, reading entire record(all bins) and then writing entire record. Record size is going to be around 8KB or 64Kb.
I’ll be using the native client, the command is given below:
/opt/aerospike/bin/asbenchmark -w RU,50,100,100 -b 10 -o I,S:100,B:1000,I,I,S:50,B:6000,B:1000,I,S:10,S:100 -latency 7,1 -r any -z 10 -h DB_IP
(The size mentioned in the above command is around 8KB as per the requirement)
Each node has got 40 core CPU(hyperthreading) and storage will be on both RAM and disk(RHEL 6).