155 result vec

[rlkelly]

I was going through some of the cleanup issues, was this the general idea of what you were looking for? It filter_maps so you don't have to iterate later

[garious]

Unfortunately, no. That PR has two issues, first the loss of error information from that early filter_map, and second, we actually want to go the other extreme and propagate Vec<Result<T>> through the entire stage so that we never need to malloc.

[rlkelly]

Ok, so still exploring how to improve this pipeline. I'm not 100% sure how to prevent malloc's in Rust.

My thought was filtering earlier because we see here: https://github.com/rlkelly/solana/blob/db019421ae44e109e560856c413f4c9ea2fd732f/src/bank.rs#L272
that this are just filtered and will return an error at process_verified_entries so would failing earlier be better?

[garious]

The bottleneck is very likely BankingStage::process_packets(). We'll need to benchmark it (which is a little tricky) and then either optimize it or make the stage shorter by moving serialization and/or deserialization to separate stages. I have some time in a couple hours to whip up that benchmark. Kudos if you beat me to it!

[garious]

Hint: see to_request_packets. That type signature should be "T: Serialize", not Request. If you had that, the benchmark should be a breeze to write.