Sun-Jc
commented
6 years ago Thanks for your comments. Let's focus on your questions about Nebulas Rank.
-
Simple answer
Cheating could be mitigated by mechanism design.
-
More details(tr;dl)
- With regards to the computation of nebulas rank, it is designed to be allowed to run on every full node, which will produce the same output, given the agreed block chain data as the input. So if any node is not willing to compute the rank itself, it could fetch the data from somewhere else, as the rank value could be verified at any time.
- About cheating on ranking, it is also the main concern of Nebulas Rank.
- First, from the prospective of incentive mechanism, Nebulas Rank is some kind of better metric than the "stake" for POS. The intuition is that POS encourages agents to own more token but not to spend any. But in order to encourage the system to develop, the owner of account should have the ability to create more liquidity, etc.(we could call it contribution). So a ranking algorithm should care about the local information of a node, such as 1-hop one, i.e. degree centrality, and \inf-hop one, i.e. centrality of page rank family.
- Second, unfortunately, degree centrality is not hard to improve in transaction network. So the "unforgeable power", i.e. stake, of an account should also be considered by ranking algorithm. In my personal opinion, in the worse case with all accounts try hard to transfer to each other and acts "normally", the ranking based incentive mechanism would be no worse than POS. So it is safe enough as long as POS-based consensus is safe. For example, by incorporating coinage factor, a wealthy and active account might earn "slow money" by Nebulas Ranks, but a wealthy and lazy account could earn quick money. However, a wealthy and lazy account should be ranked low by NR. [0,2,5 pt in your last argument paragraph]
- Third, with the above two points, the ranking algorithm faces a trade-off between stake and contribution. The main promise is, to raise its rank, an adversary must hold quite an amount of wealth in the system, and it must be influential enough and the benefit of cheating should bounded by a concave function. Besides that, more processing method makes it harder to cheat. For example, by encouragement function, an adversary has to preserve half of the income in the account each time, which reducee its influence exponentially. [3 pt in your last argument paragraph]
- As for other preprocessing methods, they could also mitigate some common cheating effect. Of course, it is not the main reason that NR could resist cheating. However, they are simple but effective according to the Ethereum transaction network analysis. [1, 4 pt in your last argument paragraph]
- The last but most important point, we are working on a better version of NR, using cooperative game theory. With the new design, all concern above should be resolve by more rigorous theorem. Sybil attack on ranking could be mitigated by the mechanism itself. Let's release more detail after peer review. And you are welcome to contribute to it :)
- By the way, cheating, or shua dan, is believed to be well resolved by mechanism design. For example, you could refer to recsys'16 .
jianhuaixie
Nebulas
Nebulas，基于价值尺度的区块链操作系统及搜索引擎。
背景：
当前区块链，一方面无法满足大规模商业应用的扩容性，另外一方面存在价值尺度缺失，各个区块链项目五花八门，技术角度不一，吸引人群有差异，宣传力度也不一致，没有一个统一的价值尺度来衡量这些区块链项目。
星云链的目标
主要内容包含：
星云链设计原则
NR确实有比较大的商业潜力，但如何对抗人性中的弱点才是亮点。如果搞出一个不错的指数，但抵不住人性的恶，分分钟被作弊薅羊毛了，这个指数有还不如没有。
将星云链上的应用简单理解成智能合约，这些智能合约是链上数据的一部分，通过链上数据的追加实现基础协议的升级。这个说法很有问题，升级智能合约是大家都想的，但如何达成共识来升级呢？星云链通过在智能合约底层存储支持状态变量可跨合约访问的设计，完成智能合约的升级，这一句感觉有点玄乎呢？
PoD算法，利用NR的价值尺度评估出对生态贡献度较高的账户，平等赋予记账资格。如果NR是可以人为参与影响，那么遏制记账权被垄断就很难说得过去。
Nebulas Rank
NR是其他算法的基础部分，三重价值尺度：
流动性 交易的频次和规模，这个就完全可以造假
传播性 资产流动的广度和深度，也完全可以造假
互操作性 信息间的互操作，具体这些操作是什么？
其实说了这么多，就是按照对每个帐号地址进行统一价值尺度的图计算处理。然而现在的图计算是中心化的计算，基本都是采用分布式计算，比如Spark来计算，这个NR的计算难道也要用贵公司的服务器来跑？
抵抗作弊
作弊手法：
NR弱化作弊的机制：
这个NR说这么多，就是简单考量地址的热度，这个跟整个项目的热度也不是绝对的对应关系。比特币转账成本这么高，让它在钱包里呆着好了，或者直接在交易平台呆着好了，但其他交易费低的就不是这样。要用一个统一的数值来衡量所有的项目，难度还是很大。至少这个NR设计看不出有这个能力。
这些基础的疑问，如果Nebulas不出来解答，白皮书下面部分是没必要看了。