You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@IanHoang and I had some discussions today where I asked whether it would be better to maximize throughput by default. This would test the limits of the system by default, and ideally encourage users to set appropriate throughput limits for their use cases, rather than relying on the preset values we currently have.
However, some users may prefer these limits to be on by default as it's easier to compare tests using the same parameters. In this case, we could always introduce a default_throughput file with these throughput values. Running tests with unlimited throughput can also produce more inconsistent results, and users would have to adapt to the new default behavior/difference in their results.
I want to use this issue to gather more opinions on whether we should change the default behavior for target-throughput in OSB.
The text was updated successfully, but these errors were encountered:
I support the idea of having OSB disable target-throughput by default. Most users seem to be using OSB in benchmarking mode (i.e. target-throughput set to 0 or disabled) and even the public nightly runs do not use target-throughput.
Users mostly use this feature when they need to match a specific throughput level, whether that's to keep some testing variables constant or to simulate throughput levels seen in production clusters. This feels similar to when we moved the changed the default testing pipeline opensearch-project/opensearch-benchmark#326
@IanHoang and I had some discussions today where I asked whether it would be better to maximize throughput by default. This would test the limits of the system by default, and ideally encourage users to set appropriate throughput limits for their use cases, rather than relying on the preset values we currently have.
However, some users may prefer these limits to be on by default as it's easier to compare tests using the same parameters. In this case, we could always introduce a default_throughput file with these throughput values. Running tests with unlimited throughput can also produce more inconsistent results, and users would have to adapt to the new default behavior/difference in their results.
I want to use this issue to gather more opinions on whether we should change the default behavior for target-throughput in OSB.
The text was updated successfully, but these errors were encountered: