This is the current state of my cluster. Do you see any abnormalities?
I have fiddled with the number of items per bulk request and the number of threads running. Now I am managing about 8000 update requests per 20 minutes. It is still very slow. I do not see the delays in the monitoring graphs.