Unable to reproduce TPCC benchmark results

Hi,

I’m trying to reproduce tpcc benchmark result based on this link: Performance Benchmarking with TPC-C | CockroachDB Docs

I have set up 4 c5d.4xlarge machines. One is for tpcc client, the other three are for the clusters. They are under the same security group with required ports opened. I basically followed all the steps mentioned in the link above. But I failed to see a good tpmc and high efficiency shown in the link:

Here is what I got:

Not sure where I got wrong, I’m using v20.2.7 and there is no error shown in the console during the tpcc run. Can someone help me?

Thanks!

It seems like something could be misconfigured. Since you’ve gotten the data loaded, you can now run with fewer warehouses from the workload generator with the --warehouses flag. Try searching for the point at which you get good latencies - if it’s very far off, you should check your hardware and load balancer configuration.

Including more info in this post will help us help you as well. Things like screenshots of the time series graphs, more context from the load generator logs, and so on.

Cheers,
Jordan

Hi Jordan!

Thank you so much for the reply! Just like what you’ve mentioned, I’m trying to decrease the warehouse number and find a cutpoint that I can achieve good latencies.

And for the times series graphs, I know that I can get that from the metric dashboard in the consoleDB. But I’m not sure how I can get the load generator logs. Are there any docs that I can refer to so that I can provide you with more details in the follow-up.

I just mean the stdout from the load generator, like you pasted a screenshot of. Maybe there would be some hints inside of those logs - for example, did the load generator start at good latencies and get worse, or was it always bad. Including the arguments you used for the load generator also helpful.

Just want to make sure that I understand you correctly. I’m using the fixtures import command for loading tpcc since I found this way faster than init. And here is the stdout from the load generator for 750 warehouses:


And based on the previous stdout from the load generator that I’ve seen, I think this one should be a pretty standard stdout indicating the load generator is acting normal. Is that correct?

Thanks!

Hi Jordan,

I’m trying to configure the load balancer based on the link here:How to use a load balancer - YouTube
I’m using 4 c5d.xlarge machines and I got three nodes distributed on 3 machines and the last machine is my tpcc client. I generate the .cfg file in the tpcc client machine with the below content and the server IPs are correct:

But when I tried to activate the balancer, it failed to connect to any nodes, while my cluster is working fine based on consoleDB. Is there any filed I put it wrong in the config file?