MegaETH About
[REVISITING THE WORLD COMPUTER]01
A New State Trie
Updating the state root of MPT is one of the most critical bottle-necks in execution clients due to the large number of random disk I/Os it incurs. To address this issue, we designed a new state trie that is highly memory & I/O-efficient. It can scale to terabytes of state data smoothly without incurring additional I/O costs, even on nodes with limited RAM.
02
Write-Optimized Storage Backend
The MDBX database is commonly used in EVM execution clients, but its high write amplification and single-writer lock result in poor write performance. To accomodate our extremely high rate of state updates, we revamped the storage backend to optimize for write performance while maintaining predictable read latencies.
03
Bytecode Compilation
Interpreting EVM smart contracts incurs significant runtime overhead, resulting in performance up to 1000 times slower than native code in some cases. To address this, we utilize just-in-time (JIT) compilation to remove the interpretation overhead and bring near bare-metal performance to the most compute-intensive Dapps.
04
Two-Pronged Parallel Execution
To fully exploit parallel execution, we adopted a unique approach that employs distinct strategies for block production & validation. The sequencer can utilize any concurrency control protocol, without being limited to deterministic methods like Block-STM. Meanwhile, full nodes leverage stateless validation to execute blocks in parallel, thereby avoiding state contention and maximizing parallel speedup.
05
Efficient State Sync
We developed a custom peer-to-peer networking protocol optimized for low-latency & high-throughput state synchronization, making near-optimal use of available network bandwidth. Combined with novel compression techniques, even low-cost nodes with limited bandwidth can stay in sync with the network at 100,000 TPS. Additionally, new nodes can join the network and catch up to the latest state quickly.
06
Streaming EVM
To achieve our 1-ms block time target, we co-designed an efficient streaming block production pipeline with our parallel execution engine. This new pipeline operates asynchronously with various system components, such as the storage backend and state trie. Consequently, it can continuously process incoming transactions and emit results in real time.
CAREER
Backend Engineer
GTE |ENGINEERING|FLEXIBLE |FULL-TIME
DevOps Engineer
GTE |ENGINEERING|FLEXIBLE |FULL-TIME
Incentivization Strategy Lead
PUMP PARTY |Remote |Full-time
Founding Engineer
HOP |ENGINEERING|REMOTE |FULL-TIME
