Xythum L2 Benchmarking and Estimations (1)

This document aims to present current development and benchmarks achieved till date by Xythum Labs.

Prepared for Internal Stakeholders | Date: 23-03-2025

SYSTEM architecture

Core innovation introduced in system architecture is using Greenfield as DA and settling on BNB L1 chain. Highly compressed batched block data is asynchronously stored on DA and its commitment is revealed on BNB L1 open to be challenged at point in time. Hosting data on Greenfield brings down execution cost and fee on L2 close to negligible (~0.00000063 USD).

• Xythum L2 is here with a promise of offering real-time Blockchain with High frequency / High TPS. The reason being able to offer final estimated 70k TPS and lowest fee typically lies in Key DA integration of Greenfield. Since Greenfield and BNB L1 are tightly integrated, Xythum is settled on BNB L1 and data stored is on Greenfield.
• Thus Xythum remains a pure optimistic rollup rather than a anyTrust rollup using DA for storage and L1 for settlement and conflict resolution.
• Apart from Greenfield integration, Xythum relies on other techniques and optimizations in its roadmap which would be scaling Xythum to hit High Tx speed.
  • Neural compressions [ IFLOW, IVPF ] over traditional Brotli and zlib compressions.
  • Usage of parallel WASM based EVM for execution.
  • Block gas limit of 140 Million same as L1 and Block time of 0.25 second.

System architecture

Core Components:

• DA Layer: Compressed block batches stored on Greenfield (13 MB blobs every 90 blocks).
• Settlement Layer: State commitments and fraud proofs anchored to BNB Chain L1.
• Execution Layer: Parallelized EVM with a 140M gas/block limit and 0.25-second block times.

Key Innovations:

1. Greenfield-BNB Synergy:
   • Tight integration between Greenfield (storage) and BNB L1 (settlement) reduces DA cost significantly lower than existing-based L2s.
     • opBNB ~0.0000002 BNB per tx
     • Xythum ~ 0.000000001045 BNB per tx
     • 200x improvement over OPBNB
   • Batched commitments are submitted asynchronously to minimize L1 congestion.
2. Compression Stack:
   • Neural Compression (IFLOW/IVPF): Outperforms Brotli/zlib with ~40% higher compression ratios for structured blockchain data.
   • 13 MB Blobs: Achieved via multi-layer compression.
3. Optimistic Rollup Design:
   • Pure optimistic model (no anyTrust assumptions).
   • Fraud challenges resolved on BNB L1 via Xythum-Challenge SDK.

Challenge Flow

Xythum is open to be challenged and the challenge conflict will always be settled on L1 BNB chain. Challenger can sync and watch chain for anomalies by running Xythum-Challenge SDK which will be having integrated DA and sandbox execution.

Challenger would be rewarded upon settling a successful challenge and securing the chain, and respected parties would be penalized accordingly.

L2 Benchmarking and Data

Context : Xythum began its L2 development in recently and this benchmarking demonstrates current L2 throughput

Note: Current development doesn't include stats from encrypted mempools and other optimizations, refer Estimated Stats section which includes project goals and future integrations.

Below benchmarks show node running for 500 blocks

• TPS achieved: 20,000
• Block Gas limit: 140 M
• Tx types: Legacy Transfers

CHAIN DATA DUMP : https://drive.google.com/drive/folders/1I9RjstPyrE9F6rzt7QVJqeEtpcn4u30L?usp=sharing

Greenfield Stats

• We were able to batch every 90 blocks with compressed data ~13MB each blob costing 0.000006 BNB per batch

Estimates from benchmarks show Xythum L2 batch would be paying ~18 BNB on Greenfield and L1 fee for every 100 Million blocks.

Final Estimates And Future Work

• Xythum Encrypted Mempool
  • Encrypted mempools are not currently integrated into our stack, and upon future integration adds a latency overhead on our chain but its negligible.
  • https://xythum.io/mev-resistant-l2-proposal.pdf - This whitepaper about MEV resistant layer on BNB authored @ Xythum states that decryption would have a latency overhead of 7 ms per decryption.
  • We estimate aggregation and batch decryption would add latency close to 20 ms although we would be encouraging a parallel execution of decryption architecture with Sequencer being run on a high end TEE/SGX based hardware.
• Neural Compression Schemes
  • We would be integrating neural based compression schemes which have proven to be a better performant than existing Brotli and zlib compressions.
  • Compression algorithms like Iflow and IVPF have caught our attention and show promising results. On successful integrations this could even cut down gas fee by a huge Factor.
• Sequencer and Node level optimizations
  • We aim to use WASM++ EVM for fastest state transitions and parallel execution.
  • Change L2 state trie structure to support high speed syncs.
  • Integrating SIMD optimizations (single-instruction multiple-data) for batch processing of similar transactions (e.g., token transfers).
  • Replace Merkle-Patricia trie with Sui-style flat storage for O(1) read/write.

---

With above mentioned considerations here is future scalability and cost estimates.

Cost Analysis Methodology

Key Metrics:

• Current TPS 20k
• Batch size: 90 blocks → 1 blob (~13 MB).
• DA cost per blob: 0.000006 BNB.
• Total blocks analyzed: 500 blocks (extrapolated to 100M blocks).
• Transaction type: Legacy transfers (21k gas equivalent).

Formulas:

1. Batches per 100M blocks = $100,000,000 / 90$ ≈ 1,111,111 batches
2. Total DA cost = 1,111,111×0.000006 = 6.67 BNB.
3. Total L1 settlement cost = 18 BNB − 6.67 BNB = 11.33 BNB.
4. Transactions per 100M blocks = 100,000,000 × 4,000 tx per block = 400,000,000,000 tx.

Gas Cost Breakdown

Operational Costs (Per 100M Blocks):

Component	Cost (BNB)	% of Total
Greenfield DA	6.67	37%
BNB L1 Settlement	11.33	63%
Total	18.00	100%

User Transaction Costs:

• DA + L1 cost per tx = $18BNB / 400,000,000,000$ = 0.000000000045 BNB / tx.
• Execution cost (L2): ~0.00000001 BNB/tx (assumed compute).
• Total cost per tx ≈ 0.000000001045 BNB (~$0.00000063 USD). [ dependent linearly on BNB price ]