🧞Why does Starkware use block packing?

What is Block Packing and how does it Give Scalability

Aug 12, 2024

GM and happy Monday frens🖖

I've seen things you people wouldn't believe... 40$ gas fee. Transaction fees greater than the transactions themselves. Thousands of dollars lost due to network congestion.

StarkWare is trying to solve this problem with block packing, a new architecture, which decreases cost of L2 blocks.

In today's newsletter:

🌏 3 must-see events in Singapore
📦 Block packing for newbies
📰 Neeeeews

🇸🇬 Meet us at Singapore!

We have 3 events in September during Token2049 in Singapore!

💠 Modular & L2 Day | Sept 17th
🔁 Restaking & Infra Day | Sept 18th
🥮 BTC Scaling & BTCfi Day | Sept 19th

We already have speakers Crestal Network, NEAR / Nuffle Labs, Across, Arcana, Manta, Movement, Metis, Scroll & Everclear!

Grab your ticket below!

Get a free ticket

Share this newsletter and earn rewards

Click below, share this newsletter with your friends and earn cool rewards!

Click to share

🤿Deep-dive: What is Block Packing and how does it Give Scalability

Starknet is a permissionless decentralized ZK-Rollup, which operates as an L2 network over Ethereum. They use the block packing to decrease cost of producing blocks and sending them to L1.

Keynote by Head of BD @ StarkWare From L2con! Watch full presentation here.

How StarkNet works right now?

Users on Starknet use the wallet to interact with apps and submit transactions. These transactions go into a centralized mempool managed by Starknet, then pass through a simple sequencer that forms blocks. Each block is sent as a job to the shared prover.

Stark proofs are efficient because their verification is logarithmic to the computation size, costing around 9 million gas on Ethereum, regardless of the proof size. Starknet aims to include as many transactions as possible, using recursive proving. This process creates a binary tree of proofs, which is ultimately condensed into a single proof sent to Ethereum.

The shared prover handles jobs from Starknet, StarkX, and other teams. Each Starknet block is one such job, which, when verified, is settled on Ethereum. Two components are sent to Ethereum: the actual proof and data availability. Think of it as proving "x = x" without revealing what "x" is.

The recursive tree structure means each Starknet block is a leaf, independently proven, and then combined into higher-level proofs until a root verification is sent to Ethereum. This method keeps gas costs roughly fixed, providing security for the entire tree for the same cost as a single job.

What is block packing?

Block packing is the architecture that decouples the gas footprint on L1 from the number of blocks on L2. Previously, we paid 200,000 gas per block on L1 just to get started. Now, with this new approach, there's no fixed cost per block.

How does block packing work?

Currently, transactions are sent to the sequencer, which forms blocks. Each block is sent to SHARP as an individual job. With block packing, if zoom in on the proving tree, each leaf (like proof A, B, C, D) actually contains its own tree of recursively proved jobs. This means each leaf in the recursive SHARP tree consists of multiple proved jobs squashed into one proof, like proof A, B, C, D.

The advantage is that all state updates from jobs A, B, C, and D are condensed into one state difference. Only this state difference is sent to L1, reducing gas costs. For example, if Alice sends Bob a million ETH in block A, Bob sends it back in block B, Alice sends it back in block C, and Bob sends it back in block D, the naive approach updates L1 four times. With block packing, these updates are condensed into a single state update showing no net change after the transactions, thus optimizing gas usage.

Benefits of block packing

In summary, there are the facts and the proof. As illustrated in the example, the facts are condensed into one state difference from multiple blocks, resulting in a lower gas footprint. This architecture decouples the gas footprint on L1 from the number of blocks on L2. Previously, each block on L1 cost 200,000 gas, but now there's no fixed cost per block.

Here's the key takeaway: this new architecture decouples L1 gas costs from L2 block numbers. Before, users paid per block and blob, which was inefficient. Now, users pay per block packing tree, enhancing scalability. Previously, blocks were produced every six minutes to avoid high gas costs, but they have a fixed block time of a few seconds without worrying about L1 costs.

This also optimizes blob utilization, spreading the cost across all transactions in multiple blocks and reducing L1 costs per transaction. This is a significant improvement, making individual transaction costs less of a burden, even with high demand. Thanks to advances like 4844, L2 scaling, and blobs, average transaction costs are now below five cents, making blockchain interactions more affordable and scalable.