2 posts tagged with "mainnet-scale"

This week, the Leios team focused on infrastructure improvements, formal methods advancement, and large-scale network simulation. The team successfully resolved outstanding CI issues, enhanced the formal specification with Full-Short Leios support, and began simulating a realistic 10,000-node pseudo-mainnet topology.

Infrastructure improvements​

• Fixed outstanding CI bugs #368 and #379, enabling all CI checks to pass.

Formal methods advancement​

• Added Full-Short Leios as a special case of Short Leios to the formal specification
• Implemented trace verification capabilities for Full-Short Leios.

Pseudo-mainnet topology simulation​

• Designed and initiated comprehensive simulations from 1 to 300 TPS using the new pseudo-mainnet topology
• Created a realistic 10,000-node network with:
  • 2,657 block producers and 7,343 relay nodes
  • Realistic stake distribution and geographic distribution
  • Two relays per block producer with realistic latencies
  • 298,756 total connections with 6-hop network diameter
• Observed significant performance challenges within the large-scale simulation:
  • Rust simulation: six minutes of network time in 10 hours at one TPS
  • Performance degradation at higher TPS rates (one minute network time in 10 hours at 300 TPS)
  • Haskell simulation requires optimization for practical large-network analysis.

Rust simulation enhancements​

• Implemented random sampling of transactions from the Leios memory pool to ensure different IBs contain different transactions when possible
• Added simulation support for the Leios variant, where IBs contain transaction references rather than full transaction bodies
• Enhanced transaction handling for high-traffic scenarios.

Analysis of conflicts and incentives​

• Completed comprehensive analysis of transaction conflicts, ledger design, and fee incentives
• Key findings on conflict management:
  • Honest duplicates and conflicts are unavoidable with local sortition
  • Memory pool rules can minimize conflicts through prompt transaction removal
  • Collateral requirements for failed transactions conflict with Cardano's guarantees
• Identified block producer compensation strategies for handling conflicting transactions
• Proposed EB-level optimization through bitmap-based transaction validation to reduce persistent storage of duplicates and conflicts.

This week, the Leios team made significant progress in simulation capabilities, with a successful comparison of Rust and Haskell simulations across 90 scenarios. A mainnet-scale analysis of Leios on a realistic 3,000-node network revealed unexpected performance benefits from network topology. Insights from sharding performance analysis provided important optimization strategies. Finally, the team refined both simulation implementations for greater realism and comparability, while the formal methods team developed initial trace verification tools for Short Leios.

Simulation comparison​

• Compared 90 scenarios between Rust and Haskell simulations at tag leios-2025w11
• Recent fixes and adjustments enabled meaningful comparison between simulations
• Identified issues requiring further investigation.

Analysis of mainnet-scale simulation​

• Completed the first analysis of Leios on a mainnet-scale network simulation using the Rust simulator
• Discovered that a 3,000-node mainnet-scale network transports IBs faster than an artificial 100-node network
• Identified 'shortcut' edges in larger networks as a likely factor in the improved transport speed.

Performance analysis of sharding​

• Created computational models to analyze the relationship between the fraction of shards without an IB and the expected number of extra IBs
• Evaluated performance characteristics of the simplest sharding scheme.

Haskell simulation​

• Fixed a bug in the relay protocol that prevented full diffusion of votes
• Adjusted the priority of certified EBs for inclusion in RBs
• Added support for an output log format that shares a common subset with the Rust simulator
• Analyzed TCP realism in comparison to idealized diffusion:
  • Discovered that higher IB rates and sizes improve diffusion times
  • Identified ledger state access as a significant source of latency.

Rust simulation​

• Expanded logs to include total IB size and parent ID of RBs
• Implemented the same EB selection strategy as in the Haskell simulation
• Added validation of IB headers before propagation to neighbors
• Investigating lower congestion in the Rust simulation compared to Haskell.

Formal methods​

• Developed the initial trace verifier for Short Leios simulation traces in leios-trace-verifier.

Research​

• Progressing on ledger design by exploring options and trade-offs
• Analyzing how concurrent input blocks in Leios create unique ledger-level challenges not present in Praos
• Evaluating approaches that balance multiple properties, including:
  • Conflict avoidance in the blockchain
  • Guaranteed fee payment for block producers
  • Transaction eligibility and inclusion speed
  • User experience regarding fee payment
• Investigating sharding-based solutions with various optimization strategies
• Planning to share more detailed findings at Leios public meeting by the end of March
• Targeting a comprehensive recommendation for implementors by the end of April.

From Short Leios to Full Leios​

• Planning the simulation roadmap for transitioning from Short Leios (currently implemented) to Full Leios
• Developing implementation guidelines for simulators to incorporate the pipeline referencing scheme specified in the papers
• Identifying key components needed to simulate the complete ledger inclusion guarantees of Full Leios.