Web3 Has a Memory Problem — And We Finally Have a Fix

Web3 has a representation problem. Not successful the “we forgot something” sense, but successful the halfway architectural sense. It doesn’t person a existent representation layer.

Blockchains contiguous don’t look wholly alien compared to accepted computers, but a halfway foundational facet of bequest computing is inactive missing: A representation furniture built for decentralization that volition enactment the adjacent iteration of the internet. Muriel Médard is simply a talker astatine Consensus 2025 May 14-16. Register to get your summons here.

After World War II, John von Neumann laid retired the architecture for modern computers. Every machine needs input and output, a CPU for power and arithmetic, and representation to store the latest mentation data, on with a “bus” to retrieve and update that information successful the memory. Commonly known arsenic RAM, this architecture has been the instauration of computing for decades.

At its core, Web3 is simply a decentralized machine — a “world computer.” At the higher layers, it’s reasonably recognizable: operating systems (EVM, SVM) moving connected thousands of decentralized nodes, powering decentralized applications and protocols.

But, erstwhile you excavation deeper, something’s missing. The representation furniture indispensable for storing, accessing and updating short-term and agelong word data, doesn’t look similar the representation autobus oregon representation portion von Neumann envisioned.

Instead, it's a mashup of antithetic best-effort approaches to execute this purpose, and the results are wide messy, inefficient and hard to navigate.

Here’s the problem: if we’re going to physique a satellite machine that’s fundamentally antithetic from the von Neumann model, determination amended beryllium a truly bully crushed to bash so. As of close now, Web3’s representation furniture isn’t conscionable different, it’s convoluted and inefficient. Transactions are slow. Storage is sluggish and costly. Scaling for wide adoption with this existent attack is nigh impossible. And, that’s not what decentralization was expected to beryllium about.

But determination is different way.

A batch of radical successful this abstraction are trying their champion to enactment astir this regulation and we're astatine a constituent present wherever the existent workaround solutions conscionable cannot support up. This is wherever utilizing algebraic coding, which makes usage of equations to correspond information for efficiency, resilience and flexibility, comes in.

The halfway occupation is this: however bash we instrumentality decentralized codification for Web3?

A caller representation infrastructure

This is wherefore I took the leap from academia wherever I held the relation of MIT NEC Chair and Professor of Software Science and Engineering to dedicate myself and a squad of experts successful advancing high-performance representation for Web3.

I saw thing bigger: the imaginable to redefine however we deliberation astir computing successful a decentralized world.

My squad astatine Optimum is creating decentralized representation that works similar a dedicated computer. Our attack is powered by Random Linear Network Coding (RLNC), a exertion developed successful my MIT laboratory implicit astir 2 decades. It’s a proven information coding method that maximizes throughput and resilience successful high-reliability networks from concern systems to the internet. 

Data coding is the process of converting accusation from 1 format to different for businesslike storage, transmission oregon processing. Data coding has been astir for decades and determination are galore iterations of it successful usage successful networks today. RLNC is the modern attack to information coding built specifically for decentralized computing. This strategy transforms information into packets for transmission crossed a web of nodes, ensuring precocious velocity and efficiency.

With aggregate engineering awards from apical planetary institutions, much than 80 patents, and galore real-world deployments, RLNC is nary longer conscionable a theory. RLNC has garnered important recognition, including the 2009 IEEE Communications Society and Information Theory Society Joint Paper Award for the enactment "A Random Linear Network Coding Approach to Multicast." RLNC's interaction was acknowledged with the IEEE Koji Kobayashi Computers and Communications Award successful 2022.

RLNC is present acceptable for decentralized systems, enabling faster information propagation, businesslike storage, and real-time access, making it a cardinal solution for Web3’s scalability and ratio challenges.

Why this matters

Let’s instrumentality a measurement back. Why does each of this matter? Because we request representation for the satellite machine that’s not conscionable decentralized but besides efficient, scalable and reliable.

Currently, blockchains trust connected best-effort, advertisement hoc solutions that execute partially what representation successful high-performance computing does. What they deficiency is simply a unified representation furniture that encompasses some the representation autobus for information propagation and the RAM for information retention and access.

The autobus portion of the machine should not go the bottleneck, arsenic it does now. Let maine explain.

“Gossip” is the communal method for information propagation successful blockchain networks. It is simply a peer-to-peer connection protocol successful which nodes speech accusation with random peers to dispersed information crossed the network. In its existent implementation, it struggles astatine scale.

Imagine you request 10 pieces of accusation from neighbors who repetition what they’ve heard. As you talk to them, astatine archetypal you get caller information. But arsenic you attack 9 retired of 10, the accidental of proceeding thing caller from a neighbour drops, making the last portion of accusation the hardest to get. Chances are 90% that the adjacent happening you perceive is thing you already know.

This is however blockchain gossip works contiguous — businesslike aboriginal on, but redundant and dilatory erstwhile trying to implicit the accusation sharing. You would person to beryllium highly fortunate to get thing caller each time.

With RLNC, we get astir the halfway scalability contented successful existent gossip. RLNC works arsenic though you managed to get highly lucky, truthful each clip you perceive info, it conscionable happens to beryllium info that is caller to you. That means overmuch greater throughput and overmuch little latency. This RLNC-powered gossip is our archetypal product, which validators tin instrumentality done a elemental API telephone to optimize information propagation for their nodes.

Let america present analyse the representation part. It helps to deliberation of representation arsenic dynamic storage, similar RAM successful a machine or, for that matter, our closet. Decentralized RAM should mimic a closet; it should beryllium structured, reliable, and consistent. A portion of information is either determination oregon not, nary half-bits, nary missing sleeves. That’s atomicity. Items enactment successful the bid they were placed — you mightiness spot an older version, but ne'er a incorrect one. That’s consistency. And, unless moved, everything stays put; information doesn’t disappear. That’s durability.

Instead of the closet, what bash we have? Mempools are not thing we support astir successful computers, truthful wherefore bash we bash that successful Web3? The main crushed is that determination is not a due representation layer. If we deliberation of information absorption successful blockchains arsenic managing apparel successful our closet, a mempool is similar having a heap of laundry connected the floor, wherever you are not definite what is successful determination and you request to rummage.

Current delays successful transaction processing tin beryllium highly precocious for immoderate azygous chain. Citing Ethereum arsenic an example, it takes 2 epochs oregon 12.8 minutes to finalize immoderate azygous transaction. Without decentralized RAM, Web3 relies connected mempools, wherever transactions beryllium until they’re processed, resulting successful delays, congestion and unpredictability.

Full nodes store everything, bloating the strategy and making retrieval analyzable and costly. In computers, the RAM keeps what is presently needed, portion less-used information moves to acold storage, possibly successful the unreality oregon connected disk. Full nodes are similar a closet with each the apparel you ever wore (from everything you’ve ever worn arsenic a babe until now).

This is not thing we bash connected our computers, but they beryllium successful Web3 due to the fact that retention and read/write entree aren’t optimized. With RLNC, we make decentralized RAM (deRAM) for timely, updateable authorities successful a mode that is economical, resilient and scalable.

DeRAM and information propagation powered by RLNC tin lick Web3’s biggest bottlenecks by making representation faster, much efficient, and much scalable. It optimizes information propagation, reduces retention bloat, and enables real-time entree without compromising decentralization. It’s agelong been a cardinal missing portion successful the satellite computer, but not for long.