Trade Matching Throughput
Trade Matching Throughput
Average Relayer Cycle Time (ms) • Max Intents per Batch with Proof Constraints
SnarkSide’s core engine—DarkMatch—is optimized for ultra-low latency, off-chain intent aggregation, and zk-verified trade settlement with on-chain finality. Unlike traditional orderbooks, DarkMatch executes encrypted user intents in a fully private, intent-matching pipeline where no party can observe or predict order flows prior to execution.
This section provides detailed performance metrics for the trade-matching pipeline, focusing on:
Relayer batch cycles
Proof generation latency
Throughput under real constraints
Scalability limits imposed by zk proving
All data was collected from live simulations and testnet deployments using reference hardware (see hardware profile at end).
Matching Pipeline Recap
Intent Submission (Client → Relayer)
Encrypted trade intent sent over P2P or relay network
Includes nullifier, collateral, size, direction, timestamp
Off-chain Matching Engine (DarkMatch)
Gathers compatible intents into symmetric bidirectional pairs
Applies funding, oracle delay window, margin rules
Matches sealed intents, generates match commitments
Batch SNARK Proof Generation
Circuit proves:
Intent validity
Symmetry (Buy == Sell)
Margin bounds satisfied
No replay or nullifier reuse
Oracle epoch match
On-chain Settlement
Proof + calldata submitted to
IntentVerifierVault states updated (or created)
Positions opened under ZK-encrypted vault UTXOs
Average Relayer Cycle Time
Intent Reception (network)
4–12 ms
Encrypted gRPC / libp2p
Matching Logic (off-chain)
7–21 ms
2–8 intents, memory-optimized tree
Proof Generation (Groth16)
740–960 ms
Per match-pair (unbatched)
Proof Generation (Batch)
1.7–2.2 s
8–12 pairs, witness flattening enabled
Transaction Submit (EVM)
1 block
Depends on chain; assumes L2 finality
→ End-to-end match execution latency: ~2.3s → User-facing intent response time: ~<50ms (pre-match state)
Max Intents per Batch
SnarkSide’s circuit design places hard constraints on how many encrypted intents can be provably matched in one batch proof. This is governed by:
Witness size (e.g. 2–3MB max)
R1CS constraint bounds (~250k–400k)
Public input gas cost
Prover RAM overhead
Current Benchmark (Groth16):
Max intents: 24 (12 matches)
Avg constraint count: 310,000
Proof size: ~230 bytes
Verifier gas cost: ~560,000
With Recursive Proofs (Halo2, WIP):
Targeted batching: 64–128 intents
Recursive folding across match blocks
Compression of 3–4 layers into a single zk proof
Expected verification cost: constant
Comparison to Traditional DEX Matching
Intent Visibility
Public
Encrypted
Matching
On-chain / sequencer
Off-chain, zk-verified
Front-running Risk
High
None
Max Orders per Block
Unbounded
zk-bounded (12–24 intents)
MEV Protection
None
Full
Latency (1 trade)
200ms (off-chain match)
~2.3s (with proof)
Latency (batched)
Varies
2–4s (batch of 8–12)
Tradeoff: Privacy and trustlessness at the cost of short latency overhead. Recursive SNARKs reduce this amortized gap over time.
Hardware Benchmark Environment
CPU: Intel i9-13900K (24-core, 5.8GHz)
RAM: 64GB DDR5
GPU: Not used (CPU-only proving)
OS: Ubuntu 22.04
Prover: SnarkJS (Groth16, WASM backend)
Node runtime: Node.js v18 LTS
Matching Engine: Rust-native, multithreaded
Future Optimizations
Recursive SNARK batching
Match/Settle/Oracle combined in 1 proof
Q4 2025
Native GPU Prover
Rust + CUDA witness generation
Q3 2025
Layered intent netting
Collapse multiple partial fills into net pair
Q2 2025
zkVM Integration (RISC-ZK)
Treat entire matching logic as zkVM program
2026+
Summary
SnarkSide’s trade matching architecture delivers highly efficient, fully encrypted intent settlement, achieving:
<1s matching latency per pair
~2.2s batch execution time for 12 matches
24-intent batch throughput under Groth16
Recursive SNARK upgrade path for >100 intents
This places SnarkSide at the frontier of private, scalable perp trading, enabling institutional-grade execution privacy without compromising verifiability or on-chain integrity.
Last updated