-
Notifications
You must be signed in to change notification settings - Fork 1.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Remote store performance run for different workloads, cluster configurations and test configurations #9790
Comments
nyc_taxis,docrep,bic-24 (Run 1)
nyc_taxis,docrep,bic-24 (Run 2)
nyc_taxis,remote,bic-24,bri-650ms (Run 1)
nyc_taxis,remote,bic-24,bri-650ms (Run 2)
Configuration 2 - Using r6g.xlarge, 3 shards 1 replica
Run Summary
Results for configuration 2so, docrep (Run 1)
so,remote,bic-8,bs-10k,ri-10s,bri-650ms (Run 1)
http_logs,docrep (Run 1)
http_logs,remote,bic-16,bs-20k,ri-10s,bri-850ms (Run 1)
nyc_taxis,docrep (Run 1)
nyc_taxis,remote,bic-16,bs-20k,ri-10s,bri-850ms (Run 1)
|
@ashking94 Thanks for adding the results, I can confirm that I am seeing similar results with the cluster and workload configuration you have mentioned in the issue. |
With 650ms buffer:
That is a ~20% jump. Running the same comparison with http_logs as well. Cluster config is 3 data (r5.xlarge)+3 master (c5.xlarge), 3 shards with 1-replica. Default workload params, 8 clients/5k bulk_size |
Running http_logs with default workload params is still showing regression on r5.xlarge/3-shard-1-replica config with 250ms and 650ms buffer interval.
Running new benchmark on r5.xlarge host with below workload params: |
For reference, I have ran performance test for 6 different scenarios to compare "segment replication + remote store" vs "document replication" in the issue #8874 . |
We have run OSB for different workloads with the different configurations as below. We are able to see consistent improvements in the throughput for remote store runs.
Configuration 1 - Using r5.large, 6 shards 1 replica
Run Summary
Appendix
Abbreviations
bic - bulk indexing client
rbi - remote translog buffer interval
bs - bulk size
ri - refresh interval
Results for configuration 1
so, docrep, bic-24 (Run 1)
so, docrep, bic-24 (Run 2)
so,remote,bic-24,ri-10s,bri-650ms (Run 1)
so,remote,bic-24,ri-10s,bri-650ms (Run 2)
http_logs,docrep,bic-24 (Run 1)
http_logs,docrep,bic-24 (Run 2)
http_logs,remote,bic-24,ri-10s,bri-650ms (Run 1)
http_logs,remote,bic-24,ri-10s,bri-650ms (Run 2)
The text was updated successfully, but these errors were encountered: