chore: ABCI methods benchmarks

app/benchmarks/README.md

@@ -0,0 +1,27 @@
+# Benchmarks
+
+This package contains benchmarks for the ABCI methods with the following transaction types:
+
+- Message send
+- IBC update client
+- PayForBlobs
+
+## How to run
+
+To run the benchmarks, run the following in the root directory:
+
+```shell
+go test -tags=bench_abci_methods -bench=<benchmark_name> app/benchmarks/benchmark_*
+```
+
+## Results
+
+The results are outlined in the [results](results.md) document.
+
+## Key takeaways
+
+We decided to softly limit the number of messages contained in a block, via introducing the `MaxPFBMessages` and `MaxNonPFBMessages`, and checking against them in prepare proposal.
+
+This way, the default block construction mechanism will only propose blocks that respect these limitations. And if a block that doesn't respect them reached consensus, it will still be accepted since this rule is not consensus breaking.
+
+As specified in [results](results.md) document, those results were generated on 16 core 48GB RAM machine, and gave us certain thresholds. However, when we run the same experiments on the recommended validator setup, 4 cores 16GB RAM, the numbers were lower. These low numbers are what we used in the limits.