Rollup 即服务挑战与机会:从多链到多 Rollup

DoraFactoryPublished on 2022-08-08Last updated on 2022-08-08

Abstract

区块链将迎来 Web3 的 multi-rollup 未来,任何人都可以通过 SDK 快速启动他们的 rollup,并以高性能和低成本在 rollup 上运行应用程序。

区块链将迎来 Web3 的 multi-rollup 未来,任何人都可以通过 SDK 快速启动他们的 rollup,并以高性能和低成本在 rollup 上运行应用程序。

Cosmos 和 Polkadot 采用多链结构的扩展方案。他们的区块链 SDK,Tendermint 和 Substrate,被许多项目用于定制自己的区块链。这些区块链使用跨链协议,如 Cosmos IBC[1]、Polkadot XCM [2] 和 bridges[3] 来进行互动。然而,这些协议很难保证高安全性,这导致了频繁的黑客窃取。因此,跨链协议没有带来预期的效果,区块链之间仍保持相对独立。

From Cosmos Network - Internet of Blockchains[4] and Getting Started · Polkadot Wiki[5]

后来,出现了一种更安全的扩展技术,称为 rollup。rollup 将第 2 层的交易压缩成一个「批次」,上传到第 1 层,并通过欺诈证明(如 Optimistic-rollup)或有效性证明(如 ZK-rollup)证明第 1 层的状态转换的有效性。由于数据的可用性和状态的有效性在第 1 层得到验证,rollup 获得了与第 1 层相同的安全水平,确保资产可以在第 1 层和第 2 层之间安全转移。

到目前为止,许多 rollup 项目,如 Arbitrum、Optimism、ZkSync 和 StarkNet 都已经在使用。除了这些通用的 rollup,还出现了一些特定应用的 rollup,包括基于 StarkEx rollup SDK[6] 的 dYdX(订单簿 DEX)和 DeversiFi(AMM DEX)等。虽然 rollup 技术还没有完全发展起来,掌握这种技术的团队也不多,但市场上对这种技术的需求仍然很强烈。

通用型和特定应用型 rollup 来自 https://l2beat.com/[7]

rollup 提供了一个独立的执行环境,具有高 TPS、低 Gas 和从第 1 层访问所有资产的能力,这有助于区块链上的应用从 DeFi 扩展到更普遍的领域,如游戏和社交网络。我们预计 rollup 将逐渐成为一种提供给 Web3 应用程序的服务,即 Rollup as a Service(Raas)。一些项目现在正朝着这个方向发展。以太坊 以 rollup 为中心的路线图[8] 和 StarkNet 的 Layer 3[9] 架构都展示了特定应用的 multi-rollup 未来。

StarkNet 的架构描述,采用了 特定应用的 multiple rollup 作为 Layer 3,来源 Medium [10]

建立 RaaS 的挑战

rollup 在提供 RaaS 方面仍然面临以下挑战。

工程方面

首先,让我们来谈一谈 rollup SDK。人们可以在 SDK 的基础上快速部署一些配置和启动 rollup。为避免重新发明车轮,开源的 rollup 是 SDK 开发的更好选择。对于 Optimistic-rollup,Arbitrum 和 Optimism 都是开源的。从 L2beat 那里,我们可以看到 Metis 和 Boba 都是在 Optimism 的代码基础上开发的。相比之下,ZK-rollup 的开源程度不高。ZkSync 发布了 v1 版的完整代码[11],但 v2 版只有合约代码[12](嵌入了 zkEVM)。StarkEx 只发布了合约代码[13],并通过一个封闭的源代码向第三方提供其他模块。StartNet 只提供 Cairo 的代码[14]。

尽管 Optimistic-rollup 系统拥有更成熟的代码库和对 EVM 更好的支持,但欺诈证明的固有特性使它们在最终性和安全性方面远远落后于 ZK-rollup 系统。ZK-rollup 第 2 层的交易在第 1 层被证明后立即完成,而 Optimistic-rollup 第 2 层的交易由于有挑战期,需要几天时间才能完成。另一方面,Optimistic-rollup 需要更多的安全假设:至少 1-out-of-N 诚实操作者来保证欺诈证明提交,有一个抗审查的第 1 层来保证欺诈证明接受。

总而言之,我们现在就可以在现有的开源代码基础上快速建立一个 Optimistic-rollup SDK,但从长远来看,ZK-rollup SDK 似乎更有吸引力。当然,除了代码库的问题,ZKVM 的设计,即 ZKP 可验证的智能合约,也是迫切需要的。目前,各种 ZKVM 解决方案正在开发中。每个解决方案的方法仍然不统一。

一图比较 ZKVM,来源:Ye Zhang’s talk「An overview of zkEVM[15] 」

性能方面

如前所述,分批交易需要以 rollup 方式发送到第 1 层,因此 rollup 的 TPS 受限于第 1 层的存储空间,也就是数据可用性[16](DA)问题。以太坊已经提出了一系列的第 1 层存储扩展解决方案,包括 EIP-4488、Proto-Danksharding 和完整的 Danksharding(目前正在寻求提案[17])。除了第 1 层的扩展,许多项目如 Celestia 和 Polygon Avail 也在尝试扩大第 2 层的存储容量。然而,这些解决方案的安全性和易用性仍需进一步研究。

EIP-4488 和 Proto-danksharding 将如何增加区块大小,来源:Vitalik’s 「Proto-Danksharding FAQ[18] 」

在 ZK-rollup 方面,TPS 还受到 ZKP 计算速度的限制。Paradigm 和 6block 在 GPU、FPGA 和 ASIC 上有不同的硬件选择来加速计算。此外,6block 比较了几种 ZKP 分布式计算的软件架构,包括矿池、证明聚合和 DIZK。ZPrize[19] ,一个即将到来的竞赛,也激励开发者找到有价值的解决方案来加速 ZKP 计算。

确保 rollup 服务的高可用性是另一个关键问题。目前市场上的 rollup 服务几乎都是中心化的,也就是说,只有特定的运营商可以向第 1 层提交批次和证明。这是一个脆弱的设计,因为 SPOF(单点故障)将很容易导致服务不可用。由于软件错误[20]和硬件故障[21],Arbitrum 已经多次遭受了数小时的停机。许多项目都在致力于去中心化 rollup 以避免 SPOF,包括 zkSync[22]、StarkNet[23]、Polygon Hermes[24]、PoVP[25] 和 taikocha.in[26]。

经济模式

RaaS 正在考虑应用一个好的经济模式。目前,服务提供商的利润主要来自于第 1 层和第 2 层之间的交易费差距,即从第 2 层收取费用作为收入,向第 1 层支付费用作为成本。Optimism 已经发行了治理代币[27],但收入仍不足以维持开支。

rollup 和它们的费用,来自 https://l2fees.info/[28]

现有的大多数 rollup 是建立在区块链上的第三方服务,所以它们的主要收入仅仅来自于交易费。然而,我们可以跳出这种思维模式,将 rollup 视为区块链提供的原生服务。就像 Cosmos 和 Polkadot 的设计一样,整个系统包含一个区块链和附着在区块链上的多个 rollup,形成一个具有无限可扩展性的去中心化网络。通过这种方式,网络可以用相同的原生代币奖励第 1 层区块链验证者和第 2 层 rollup 运营商。这个想法类似于 Polynya 提出的 shrined rollup[29],值得进一步研究。

功能方面

与 Cosmos 和 Polkadot 的跨链协议一样,当一个区块链上部署了多个 rollup 时,cross-rollup 协议是必要的。用户也可以从第 1 层提取他们的资产,并将其存入另一个 rollup,但这个过程需要在第 1 层支付额外的费用和更多的操作步骤。一些第三方 cross-rollup 桥[30]利用流动性池来帮助用户在 rollup 之间即时转移,但这些桥和跨链桥一样容易被窃取。

Vitalik 在《终局》[31]中描述的一个未来区块链架构,当中有 multiple rollup 和 cross-rollup 桥

理想情况下,区块链应该提供一个原生的 cross-rollup 桥,由其验证者维护以保证安全。此外,这样的桥最好支持从一个 rollup 到另一个 rollup 的同步消息调用,也就是说,一个 rollup 上的用户可以直接调用另一个 rollup 上的合约。这将最大限度地提高 multi-rollup 架构中的用户体验。这里的底层技术很复杂,但我们期待着它的出现。

结语

这篇文章描述了 RaaS,即为 DApps 提供 rollup 服务。显然,区块链将迎来 Web3 的 multi-rollup 未来。任何人都可以通过 SDK 快速启动他们的 rollup,并以高性能和低成本在 rollup 上运行应用程序。在讨论了 RaaS 可能面临的所有挑战后,我们最终提出了原生 rollup 的想法,这将有助于区块链用其原生代币奖励 rollup 验证者,并提供一个由其验证者维护的 cross-rollup 桥。我们计划进一步仔细研究,并在未来的文章中详尽阐述。

Related Reads

Glue Finance Founder: ETH Has Entered a Phase of Non-Consensus, an Inflection Point Is Coming

"Glue Finance founder argues Ethereum's current price underperformance stems from its 'unfinished machine' status. Despite record usage and settlement value in 2026, ETH remains below its 2021 peak because the protocol's continued development creates a dependency on the Ethereum Foundation for guidance and fixes. Key issues include centralized L2 sequencers, governance-based freezes (as seen with Arbitrum), state bloat, a vulnerable public mempool, and looming quantum computing threats. This 'dependency discount' prevents ETH from accruing the 'ossification premium' that markets award to immutable, trustless systems like Bitcoin. The author, a self-described Ethereum maximalist, rejects two flawed escape paths: a 'war mode' shift towards centralization for speed (which would sacrifice Ethereum's core value), or simply replacing the EF with another governing body. The only solution is the 'Manhattan Plan' or 'Lean Ethereum': a concerted, accelerated effort to complete and then *freeze* the protocol's neutral core. This involves finalizing critical upgrades in consensus (Lean Consensus), scaling (targeting 1 trillion gas/sec), quantum resistance (leanXMSS signatures), and full ZK-provable execution. The goal is to pass the 'walk-away test'—where Ethereum could run forever, neutrally, without the EF. Success would transform Ethereum into the first programmable, quantum-resistant, immutable global settlement layer, flipping its current discount into a unique 'eternity premium' that surpasses Bitcoin's. Failure—stagnation or trading neutrality for speed—would relegate it to being a slower, less trustworthy competitor."

marsbit13m ago

Glue Finance Founder: ETH Has Entered a Phase of Non-Consensus, an Inflection Point Is Coming

marsbit13m ago

Glue Finance Founder: ETH Has Entered a Phase of Non-Consensus, an Inflection Point Is Approaching

Glue Finance founder argues that Ethereum's current price underperformance, despite high on-chain activity, stems from its incomplete state and lingering dependency on central stewards like the Ethereum Foundation. The core thesis is that the market is discounting ETH not for lack of use, but because the protocol remains a "machine under construction." Key unresolved issues include centralized L2 sequencers with limited escape hatches, state bloat, vulnerable public mempools, and the looming quantum computing threat. This perpetual "work-in-progress" status forces reliance on a small group of maintainers, undermining the network's promise of credible neutrality and immutability. The author, a self-described Ethereum maximalist, rejects two flawed paths: a "wartime mode" of centralizing for speed (surrendering Ethereum's unique value) or merely replacing the Foundation with another governing entity. The only solution is to complete and then "freeze" the protocol's neutral core through a focused "Manhattan Project" dubbed "Lean Ethereum." This project aims to bundle critical upgrades—consensus layer overhaul, massive scaling via ZK-proofs, quantum resistance, and statelessness—into a decisive push to finalize the base layer. Once the core rules are cryptographically solidified and beyond anyone's control (passing the "walk-away test"), Ethereum would shed its dependency discount and earn a "rigidity premium" for its credible neutrality and programmability, potentially surpassing Bitcoin's valuation. The race is between completing this hardening and the risks of protocol capture or stagnation.

链捕手14m ago

Glue Finance Founder: ETH Has Entered a Phase of Non-Consensus, an Inflection Point Is Approaching

链捕手14m ago

Hackers Steal Nearly $17 Million in 40 Days as 'Zombie Contracts' Become Their ATMs

According to an analysis published by ZeroDrift on June 22, 2026, attackers have stolen approximately $16.9 million over 40 days from five deprecated but still operational smart contracts across various blockchains. The primary issue is not a specific vulnerability but the incomplete decommissioning of legacy contracts. These "zombie contracts" often retain economic value, operational permissions, and callable functions, making them prime targets long after teams cease active development. The most significant loss occurred at DxSale, where an old locker contract lost about $7.3 million due to a forgotten control path becoming accessible again. Other affected projects include TrustedVolumes (~$5.87M), Raydium's legacy AMM pool (~$1.34M), Aztec Connect (~$2.28M), and Huma Finance V1 pool (~$101k). These incidents involved diverse systems—RFQ settlement, credit pools, liquidity lockers, AMMs—demonstrating the widespread nature of the risk. The analysis highlights that automated tools are lowering the cost for attackers to systematically scan for these long-tail targets, which have public code and weaker monitoring. In contrast, defensive practices for contract retirement remain underdeveloped. While the DeFi industry has mature audit processes for new deployments, it lacks strict protocols for securely sunsetting old contracts, which only become truly "retired" after all funds, permissions, authorizations, and trust assumptions are removed.

marsbit1h ago

Hackers Steal Nearly $17 Million in 40 Days as 'Zombie Contracts' Become Their ATMs

marsbit1h ago

Valuation Rout of Old Titans: The Demise of a Generation's Asset Valuation Framework

"The Old Titans' Valuation Collapse: The Death of an Era's Valuation Framework" Between Alibaba's 2014 NYSE debut at $93.89 and its 2026 price of ~$95, twelve years have passed with zero price appreciation. This stagnation symbolizes a wholesale valuation reset for an entire generation of Chinese internet assets. Companies like Tencent, Pinduoduo, Meituan, Bilibili, and Kuaishou have seen catastrophic declines of 80-98% from their peaks. The core question arises: what framework now prices these companies, or has the framework itself expired? The valuation logic for Chinese internet stocks followed a clear "anchor-setting and anchor-removing" process. From 2014-2017, the dominant narrative was "US comparable discounting" – applying a growth premium and governance discount to US peers' multiples. This anchor loosened with the 2018 US-China trade war and the VIE structure risk, then was violently uprooted by the 2020-2021 regulatory crackdowns (Ant Group, Didi, anti-monopoly fines). The 2022 delisting panic and subsequent 2025-2026 geopolitical shocks (US military lists, AI espionage accusations) completed the demolition. The old "US对标打折" model is dead. However, this is not solely a China story. A structural mirror exists in US "old titan" stocks ("老登股"). In 2026, even Microsoft – with robust fundamentals – saw its PE compress from a 34x median to 22x, its worst performer status among the "Magnificent Seven" driven by a $190 billion annual AI capex crushing free cash flow. The core dilemma is universal: legacy platform giants, whether Alibaba or Microsoft, are spending colossal sums to chase an AI paradigm that may颠覆 their own high-margin, user/subscription-based business models. They have shifted from "companies defining the future" to "companies needing to prove they won't be淘汰ed by the future." This phenomenon of a dying valuation坐标系 has a historical precedent: post-1989 Japan. After its bubble burst, the "Japan premium" narrative ("most efficient manufacturing + perpetual growth") collapsed. A 25-year valuation vacuum ensued until Warren Buffett provided a new language in the 2010s: "low valuation + high dividend + governance reform." China's internet sector is now in a similar vacuum six years into its reset. While different from Japan's deflationary context, the parallel is clear: the old macro assumption of "deep integration with global capital" is falsified, but a new pricing framework is absent. Potential "new languages" for Chinese internet valuations are contradictory. AI transformation requires gutting profitable core businesses (e.g., Alibaba's ad-driven e-commerce) for an unproven consumption-based model, risking a Microsoft-like cash flow crunch. Alternatively, shareholder returns (buybacks/dividends) could build a floor, following Buffett's Japanese playbook, but current scales are insufficient to form a standalone anchor. The current state mirrors mid-1990s Japan: the old framework is dead, the new one unborn. The market waits in a vacuum for a重新定义ing force – a person, event, or proven business model shift – to answer "why buy." This may only be the middle phase of a prolonged re-rating.

marsbit1h ago

Valuation Rout of Old Titans: The Demise of a Generation's Asset Valuation Framework

marsbit1h ago

Trading

Spot
活动图片