Arbitrum day slides

--- type: slide slideOptions: theme: solarized --- ### Just a smol coredev living in an L2 World ### arbitrum day 2023 <img src="https://hackmd.io/_uploads/SJnP3ZVK3.png" width="300" /> #### *Terence Tsao* #### Core developer @ Offchain Labs ---  ## Agenda - My journey from L1 -> L2 - L1 and L2 tradeoffs - Ethereum - Arbitrum - L1 supercharges L2 - Common fallacy in thinking about L2 - L2 forward-looking ---  ## My little journey ![](https://hackmd.io/_uploads/H1F8VsrF2.png) ---  ## My time spent ```mermaid pie title "L1" : 60 "L2" : 20 "MEV": 20 ``` --- ## L1 versus L2 ---  ## L1 Consensus Actors - Validators - 700k active, 50k pending - 32eth / validator - ~20% of the ETH is staked - Nodes - CL + EL clients - 5000 nodes - free to run ---  ## L1 Consensus in a nutshell - LMD-GHOST - Favors liveness - New block every 12s - 50% online -> liveness - FFG - Favors safety - New finalization every 6.4mins - 66% online -> safety - One validator, one vote, every epoch - 1800 msgs/sec (overhead!) ---  ## L1 in action ![](https://hackmd.io/_uploads/rJlhFpSYh.png) ---  ## Decentralization has a cost - 6 consensus clients - 6 execution clients - Rough governance - 1-2 big EIPs, 5-10 small EIPs / year --- ## Staking distribution ![](https://hackmd.io/_uploads/ryqzWR5t2.png) [beaconcha.in](https://beaconcha.in/pools#distribution) ---  ## L2 Consensus Actors - Sequencer - Order txs and emit a feed - Cannot steal - Centralized, decentralize later - Proposer / Validators - Propose L2 block using feed - Can steal - Permissioned, permissionless later - Challenge game: honest party can **always** win, users can **always** safely exit to L1 ---  ## Arbitrum in action ![](https://hackmd.io/_uploads/Hk5JDSidn.png) [diagram](https://developer.arbitrum.io/inside-arbitrum-nitro/#sequencing-followed-by-deterministic-execution) ---  ## L2 Consensus in a nutshell - Sequencer - Ordering policy: - FCFS - Time-boost (soon) - Centralized - Participant count: 1 - Liveness assumption: 100% - Encryption/decryption committee - Participant count: 1-20 - Liveness assumption 33%-50% - Validators - Fraud prove - Safety threshold: 1 out of N ---  ## Decentralize where it matters | | L1 validators | L2 sequencer | L2 validator | | ---- | -------- | --- | -------- | | multiple implementations |yes|nice to have|no| | multiple node runners |yes|yes|one active, one defensive| --- ## It's all about trust ![](https://hackmd.io/_uploads/HkhExWuth.png) https://vitalik.ca/general/2020/08/20/trust.html ---  ## Leverage L1. Don't copy L1 - L1 is fundamentally very different than L2 - If something works well on L1, doesn't mean it'll work well on L2. Some ideas that don't make sense over time: - PBS - Front-running / toxic MEV - Got its principal-agent problems - Base layer sequencing - Slow without pre-confirmation --- ## L1 at time looks like ![](https://hackmd.io/_uploads/BJme6j_Fn.png) --- ## We hope L2 to look like ![](https://hackmd.io/_uploads/ryZ8JnOFn.png) --- ## Why I am bullish on L2 --- ## L1 has to make certrain tradeoffs <img src="https://hackmd.io/_uploads/BJj2UIhYh.png" width="800" /> --- <img src="https://hackmd.io/_uploads/r1KNykkc2.png" width="500" /> ---  ## L1 supercharges L2 in three areas - Scalability - Security - Decentralization ---  ## Data sharding - EIP4844 (Q4/2023) - Call data -> Blob data - Target: 32kB/s Max: 64kB/s - L1 data cost reduction 3-10x - Danksharding (in 3-5 years) - Super blob data via data availability sampling - Target: 170kB/s Max: 340 kB/s ---  ## Ethereum Proof of stake security - PoW -> PoS - Reorg can be "seen" on-chain - Economic finality - Validators are attributable - Superior censorship resistance properties ---  ## Faster confirmation rule - Heuristic on the block will remain canonical under synchronous - Break when the network does not remain synchronous - No accountability, a confirmed block can be reorg, and no one is slashed - Assume adversary < $1/3$ of network - < 1min latency to confirm - [Research post](https://www.adiasg.me/confirmation-rule-for-ethereum/) ---  ## L1 on-chain censorship racle - Heuristic on whether there has been censorship - Assume % validators are honest - Count skip slots over $N$ period using a cumulative distribution function - $Pr(X≤k)=∑ki=0(ni)pi(1−p)n−i$ - Less than 34 missing blocks out of 688 slots (2hrs) - Confidence $10^-6$ - Equally less than 4 missing blocks out of 225 slots (45mins) - [Research post](https://ethresear.ch/t/reducing-challenge-times-in-rollups/14997) ---  ## Single-slot finality - Long-term goal, may not ship until 2025 - Finalize every 12s - Networking and message aggregation challenges - [Research post](https://notes.ethereum.org/@vbuterin/single_slot_finality) ---  ## Decentralization: statelessness - Verkle trie ("Vector commitment" and "Merkle Trees) - New data structure to reduce data to validate - Validate data using witnesses. Under polynomial commitment, witness size is small and can be gossip over p2p - Two testnets are up and running ---  ## Decentralization: bounded historical data - EIP4444: Stop serving historical data over p2p and locally prune historical data - Historical data is not necessary for validating new blocks - Historical blocks & receipts occupy more than 400GB of disk space (and growing fast!) - To validate the chain, nodes must have a 1TB disk ---  ## L2 in the not-so-far future - Scalability - Cost reduction 3-10x - Security - Faster L1->L2 messaging - 1min to deposit - Faster L2->L1 messaging - 1-2hrs to withdrawal - Decentralization - Running a L2 node becomes more accessible ---  ## L2 open innovations - EVM+ (Stylus) - Privacy - Decentralized sequencing - Permissionless challenge game - L3 --- <img src="https://hackmd.io/_uploads/H1VzXkych.png" width="500" /> ---  ## Call for actions - What can L1 do more? - What can L2/L3 do more? - What will you build? --- ## Thank you! ### Enjoy the rest of the arbitrum day! @terencechain <img src="https://hackmd.io/_uploads/SJnP3ZVK3.png" width="300" />