Scaling Transaction Processing

The InfiniSVM transaction processing architecture is designed for high throughput and low-latency execution. It employs a multi-stage pipeline that optimizes transaction verification, execution, and state updates using parallel processing, RDMA, and FPGA acceleration.

​

Transaction Processing Pipeline

Each transaction follows a structured pipeline to ensure efficient execution and consistency:

1. Ingress & Verification

   • Transactions enter the network via an initial ingress point.
   • Signature verification (sigverify) and local deduplication are performed to filter redundant transactions.

2. Pre-Execution

   • Transactions undergo pre-execution to determine their state dependencies.
   • Intermediate transaction effects and snapshots are sent to the sequencer via InfiniBand for efficient state tracking.

3. Sequencer & Execution Path Selection

   • The sequencer, utilizing an SDN switch and FPGA acceleration, determines execution paths:
     • Simple Path: If all accounts involved are at their latest state, the transaction bypasses additional validation.
     • Complex Path: If at least one account has a newer version, the transaction enters a local mempool for ordering and re-execution.

4. Parallel Execution & State Updates

   • RDMA applies state changes directly for simple path transactions, leveraging a local cache on the SDN switch.
   • Complex path transactions are scheduled in the local mempool, ensuring fairness and optimal parallel execution.

5. Finalization & Propagation

   • Once execution is complete and state changes are committed, the transaction is broadcast globally through a Proof of Propagation (PoPs) network.

​

Optimized Scheduling & State Storage

To achieve sub-millisecond processing speeds, the sequencer:

• Implements an optimized scheduling algorithm to parallelize dependent transactions.
• Stores account data in a distributed database, ensuring scalability and data integrity.

The InfiniSVM architecture integrates SDN, RDMA, and FPGA-accelerated sequencing to enable high-performance transaction processing. By intelligently routing transactions through simple or complex paths, it minimizes latency while maximizing parallel execution efficiency.