I have a 5 node cluster. When I perform a load test with 100 virtual users which uses a Redis cluster, I receive an average response time of 300-400ms. What approaches can I make to Redis to reduce the latency down to 100ms?
@matt please provide a bit more details so that we can guide you through the improvement of command latency process.
- which command’s performance are we talking about
- what is the data size
- whats the read/write pattern
- have you confirmed that the DB processes are fully saturated in CPU?
- have you confirmed that the client is not the bottleneck?
- have you confirmed the expected network packet latency between the two/more VMs involved in this benchmark? likewise, have you measured the maximum PPS and BW of network?
- which benchmark client are we discussing?
Hi @fcosta_oliveira, I was able to identify the bottleneck which was due to a high number of reads and writes to the same keys at the same time. Changing the approach for reading/writing keys to Redis resolved the issue for me.