SwingSet benchmarking tool, phase 1 #8239

…ark stuff - Change the default parity of init vs. resume: The `--init` command line option is now redundant as it is now the default (though it is still supported for backwards compatibility). In its place is a new `--resume` option with the same meaning of the older (also still supported) `--noinit` option. What this means is that absent command line parameters to the contrary, the default behavior of a swingset-runner execution will be to re-initialize the kernel state at the start of a run. This is a UX simplification, a reflection of the fact that in practice we found we wanted `--init` about 99% of the time, while exercising swingset resumption was very rare. Various scripts that depended on the old default behavior have been updated for the new world order. - There is a new command line option `--usebundlecache` that if present will configure the swingset being run to use a bundle cache. This will noticeably speed up successive executions of swingsets to a lot of bundling (for example, if the have a lot of vats). The bundle cache will be placed in the same directory with the swingstore (which is also typically where the swingset's source files and config file (if it has one) are maintained).

Remove awaits that were blocking bootstrap from completing because they were waiting on events that can only happen after bootstrap is done, while taking care to log any rejections that these post-bootstrap activities yield.

…m bootstrap

In this mode, the configured swingset is loaded indirectly, interposing a swingset-runner-provided benchmark controller as the swingset's bootstrap vat. The benchmark author is expected to provide a benchmark driver vat in a file named `vat-benchmark.js` adjacent to the swingset's config file. The benchmark driver vat is expected to expose the methods `setup` and `runBenchmarkRound`. The controller bootstrap invokes the swingset's "real" bootstrap and then tells the benchmark driver vat to perform setup for the benchmark. It then serves as a controller for triggering the execution of the directed number of bootstrap rounds. See demo/vaultPerfTest/vat-benchmark.js for an example.

This mode is invoked via the `--chain` command line option. It adds bridge and timer devices, simulated chain storage, and kicks off bootstrap using the core proposals manifest mechanism. This is an attempt to provide a minimal emulation of the services provided by cosmic-swingset without actually putting cosmic-swingset itself into the mix. I expect the functionality here to evolve considerably as we add more chain benchmarks.

Note: This commit also includes the stripped version of the Ava-based vaults benchmark, which is like the regular Ava-based benchmark but has had all the timing and data collection code removed from it (making it much shorter and easier to read) in the expectation of external rather than internal measurement. This was an engineering stepping stone to getting the benchmark to run inside a swingset, but is being committed to the repo because we suspect it may be a useful reference in the future.

…n context

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

SwingSet benchmarking tool, phase 1 #8239

SwingSet benchmarking tool, phase 1 #8239

Commits on Aug 29, 2023