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
Most recent version of wasmtime has a new incremental compilation cache which might significantly improve our cold startup times; quoting the changelog:
An incremental compilation cache for Cranelift has been added which can be enabled with Config::enable_incremental_compilation, and this option is disabled by default for now. The incremental compilation cache has been measured to improve compile times for cold uncached modules as well due to some wasm modules having similar-enough functions internally.
We should check whether this could help us, and if so how much exactly. A good way to do this might possibly be to e.g. take the last few runtime blobs used on something like Kusama, and then write a benchmark which just instantiates them one after another, and then just compare a run where this new incremental cache is disabled with a run where it is enabled and compare the total execution time and the instantiation latencies.
This could also be helpful for PVFs; I'm guessing we could probably benchmark that in a similar way?
Most recent version of
wasmtime
has a new incremental compilation cache which might significantly improve our cold startup times; quoting the changelog:Related issue where we see that cold startup time is a problem: paritytech/polkadot#5821
We should check whether this could help us, and if so how much exactly. A good way to do this might possibly be to e.g. take the last few runtime blobs used on something like Kusama, and then write a benchmark which just instantiates them one after another, and then just compare a run where this new incremental cache is disabled with a run where it is enabled and compare the total execution time and the instantiation latencies.
This could also be helpful for PVFs; I'm guessing we could probably benchmark that in a similar way?
cc @pepyakin
The text was updated successfully, but these errors were encountered: