search

consensus-shipyard / ipc

🌳 Spawn multi-level trees of customized, scalable, EVM-compatible networks with IPC. L2++ powered by FVM, Wasm, libp2p, IPFS/IPLD, and CometBFT.
https://ipc.space
Apache License 2.0
41 stars 35 forks source link

Reprice IPLD and extern costs #974

Open maciejwitowski opened 4 months ago

maciejwitowski commented 4 months ago

From @raulk

The IPLD and extern costs were calculated based on Lotus+FVM benchmarks, and they both suffer from FFI penalties. Given that ref-fvm running inside Fendermint has no FFI boundary to cross, we should benchmark and reprice the IPLD syscalls and the externs. This should increase the chain throughput when the system is saturated.

maciejwitowski commented 4 months ago

From @aakoshh

We haven’t implemented any limits on how many transactions can be put into a block apart from the default byte size in CometBFT so I don’t think this currently affects the benchmarks

From @raulk

Interesting. We'll need to add block gas limits in that case. There is no correlation between serialized tx size and computation cost. So in the current form, one could easily slow down a subnet by submitting tiny transactions with large computational footprint each, if one is willing to expend the gas.