原文：Variable security data sharding —— polynya

Translator: Evelyn｜W3.Hitchhiker

Disclaimer: Everything here is purely speculative thinking, there are many oversimplifications, and none of this should be taken seriously. I just hope it provides some food for thought.

The beautiful idea of ​​Danksharding is this: only builders (which is an honest minority assumption anyway) will need to run expensive hardware. Over time, rollups will scale to millions of TPS, with only very small costs incurred by validators, users, and everyone else. The problem is, this ambitious vision will take time to work. Probably anywhere between 2 and 5 years, depending on who I ask. While crypto has always suffered from extreme planning fallacies, things are certainly improving, but I refuse to believe any roadmap until I can see a fully functional prototype.

EIP-4844 is a huge step forward, and I believe it will be more than sufficient for all applications and valuable transactions that require high security for the foreseeable future, and will enable rollups to scale 100x-1000x over current activity. However, there will be different classes of data-hungry applications that may require "millions of TPS" and not require high security.

At the same time, it’s clear that developers aren’t planning on waiting for full danksharding. Today, 7 of the top 15 projects on L2Beat are not rollups, but validiums and optimistic chains that use external data availability layers. (Here, I’m assuming you’re familiar with these constructions and why they’re still superior to sidechains, etc.) StarkEx (now with 6 validiums!), Arbitrum, and Metis already have external data availability layers; while StarkNet, zkSync, Polygon, and others are all building internal data availability layers, not to mention EigenDA or Celestia, etc. These hybrid solutions require neither EIP-4844 nor full danksharding to keep transaction fees low, and the key is that they’re already here.

One approach is to let them do their thing, with users choosing less secure validium options (but still higher than alt-L1s) when they need them. The magic of Volitions is that users can choose this on a per-transaction or per-user basis.

But another way to look at it is: how can we improve on the external data availability layer?

While this is far from a representative sample, today 20% of Ethereum validators exceed the norm. Meanwhile, 45% like to have very conservative bandwidth/traffic requirements. With EIP-4844, we are optimizing for this 45%, but the remaining 55% will continue to run on underutilized bandwidth. So the idea is about taking advantage of this idle bandwidth by adding simple data sharding, where the validator set will be split up. I know nothing about engineering, but I take Dankrad's word that it's "trivial".

With EIP-4844, we have a new data availability layer. I call it Data Shard 0 (DS0). DS0 is mandatory and guaranteed by the full set of Ethereum validators.

Here we can have more data shards (S1, DS2, ... etc) that can opt-in from validators. So, 30% of the above 2TB-10TB camp can run 2 or 3 data shards. So while DS0 is secured by 100% of the Ethereum validator set, DS1 is 55%, DS2 is 50%, and so on. Now that we have those over-spec validators who are comfortable with 20TB+ traffic or 100Mbps+ bandwidth (who apparently mistakenly think they are Solana validators) running, that 20% can also run more data shards. So, your last data shard (DS16) only provides 10% of Ethereum security. Of course, I make it sound arbitrary, but depending on the structure of the Beacon Chain committee, there may be some clear divisions. But the gist of it is that DS0 comes with 100% security (so all rollups will be HRE). DS1 has 50% security, DS10 has 25% security, DS16 has 10% security, and so on, for a new construction type that is between a full rollup (DS0 / 100% security) and a validium / optimistic chain.

In reality, as we see many validiums and data layers coming online, the security required for each application, user, and use case is a spectrum. Not everything needs to be secured by hundreds of billions of dollars of economic security.

But here’s the thing, if we assume 30% of ETH supply is staked, even a “minimum security” DS16, even in this bear market, is still backed by $5 billion in economic security, and that’s still a top alt-L1 tier security margin. It’s also important to remember that this security margin is only for data availability. Proofs of validity and fraud proofs are still verified at 100% security! So even a half-rollup/half-validium of a DS16 has significantly higher security than a top alt-L1.

This may still be a significant improvement over external data layer options. To be sure, high-value financial transactions will settle on DS0, but some medium-value NFTs may only need DS8, and for zero-value gaming applications, even DS16 may be overkill. As mentioned above, in a voluntary type of setting, each application can choose the level of security it needs based on its own circumstances.

What about rollup users? They only need to run the data shards that are relevant to them, so the system requirements will not be different from EIP-4844. This is why even if there is a security compromise, there is no decentralization compromise relative to EIP-4844. (Unless it is a gigarollup settled on multiple data shards - but this is something its users will be well aware of)

To be clear, I don't expect any of this to happen, these are just the ramblings of a crazy non-technical amateur blogger. The path of least resistance is EIP-4844 arriving in 2023, and once it saturates in a few years, data-hungry use cases will spread to many external data layers, and data will all be highly commoditized and abundant. Eventually, over the years, danksharding will gradually come online and become the best solution. But maybe someone will read this and find a better idea to fill the huge gap between 4844 and full sharding...