Vision #1
If we are to build new financial foundation for the whole world, the system must be decentralized and
censorship-resistant.
Vision #2
If we are to digitize all assets in the world, we need horizontally scalable solution.
Vision #3
If we are to commoditize blockchains, we must achieve web2-like reliability and flexible architecture
to support app-chains.
Steps
- System that scales to millions of nodes
- Consensus with sub-second finality
- Flexibility to support app-specific chain
News
Yes. Avalanche solves scalability while decentralized with novel consensus algorithm and subnets.
What is Avalanche?
Overview
Avalanche ($AVAX)
- Mainnet launched in Sep 2020
- Supports Ethereum Virtual Machine (EVM)
- Novel Consensus Algorithm: Snowman
- Proof-of-Stake
- Fast and Scalable L1 (>6K TPS)
- Reliable (no downtime, no reorg)
- Platform for deploying "Subnets"
Avalanche Today
>3.6B requests (a day, July 2022) with ~80ms latency
Avalanche Today
Subnet horizontally scales!
>500 TPS (possible >6,000 TPS)
Avalanche TPS
Customize subnet-evm to get super-high TPS!
Avalanche Consensus + Subnets
- Consensus with sub-second finality (Fast)
- Decentralized with >1,200 validators (Secure)
- Subnets for application specific chains (Isolation)
Avalanche Consensus ☃️
Snowman Protocol
Consensus
"Assume a collection of processes that can propose values. A consensus algorithm ensures that a
single one among the proposed values is chosen." Leslie Lamport, Paxos Made Simple
(2001)
Should this transaction be placed in a block or not?
PoW or PoS is NOT a consensus mechanism!
Consensus Until Now
Classical (Lamport 1998,
Paxos/Raft/etcd)
- Quick finality but does not scale
- Quadratic message complexity
- Permissioned, requires precise membership
Consensus Until Now
- Robust, no need for precise membership
- High latency, low throughput
- Wastes energy, not green, not sustainable
Avalanche Consensus Family
- Published in 2020
- Instant finality, low latency (~1 sec)
- High throughput (>6,000 TPS)
- Scales >10-million nodes
- Robust, no need for precise membership
- Leaderless
- Quiescent, green, sustainable
- Inspired by epidemic protocols and gossip networks
- New idea: deliberately metastable
Avalanche Sustainability
Binary Consensus
- Pick one red/blue -- no correct
answer
- Adopt the majority color by repeated sub-sampling
- Consensus results in the entire network agreeing on either red
or blue
- Even with 50/50 split, random perturbation in the sampling results in a single value being
selected
At the beginning, pick any color (no correct answer)
Radomly sub-sample the network
"Red" is the majority from the sample
Adopt the majority color, "red"
Repeat this random sampling in parallel, in all nodes
Repeated random sampling perturbs conflicting state
Sequence of metastable process of random sampling
All converge to the same value (agreement)
Fast Finality
Avalanche == no block re-org, no longest chain rule
- Re-org in game? May lose item after tx revert
- Re-org in bridge? Must wait longer for source chain
- Re-org in payment? May lose money for goods
Avalanche Subnet 🔺
What is Subnet?
Primary Network == Special Subnet
- X-chain runs on DAG, used for exchanging assets
- P-chain coordinates validators and subnets
- C-chain executes EVM contracts with ETH RPCs
Subnet validator must validate primary network!
Subnet (sub-network)
Subnets
Subnets provide failure isolation
Failure in one subnet does not impact the other!
Thus better fault tolerance and better reliability!
Subnet (sub-network)
Custom networks running on Avalanche
- Security: Choose who and how many can participate
- Compliance: Comply with specific industry, jurisdiction, regulatory environment (KYC)
- Custom Execution: Common VM (subnet-evm with custom gas token), custom VM
optimized for own use case (key-value store, gaming, Rust 🦀)
- Privacy: Controls data visibility (encryption)
Effects of Subnet
Avalanche vs. Cosmos/Polygon(PoS)/Klaytn
No upper cap on consensus membership.
[1] Theoretically, Avalanche can scale beyond millions.
Avalanche vs. Cosmos/Polygon(PoS)/Klaytn
Sub-second finality, no block re-org (합의 취소).
[1] Blocktimes are configurable for faster finality.
[2] The chain halts to prevent re-orgs.
Avalanche subnets vs. Layer-2
🔥 Shared security is not a desired property 🔥
Instead, subnet provides elastic/on-demand security.
Why validate subnet?
Subnet has its own token and rewards.
Transfer between subnets?
Yes, possible (soon).
Rust 🦀 virtual machine?
Yes, possible (today, tweet).
Contributions
- Discussed why we need blockchain
- Discussed why blockchain must be scalable
- Explained why Avalanche blockchain is scalable
- Reliability with the fastest tx finality
- Flexibility to run custom VMs with custom tokens