27 May, 32 tweets, 10 min read
1/ Why #Avalanche is even better than I initially thought (part 2)?
2/ In our previous thread we discussed how *fast* the distributed #Avalanche *consensus* mechanism can sync all honest nodes:
3/ Above you see how after *only* 3 rounds the entire set of honest participants are in sync: Despite 15% being faulty (or malicious), the system manages to achieve the max. possible consensus level of 85% (for the overall network).
4/ Please note, that we performed here 1000 simulations for 10000 nodes using the *current* stake distribution. Synchronizing honest nodes is really fast!
5/ Let's experiment with another distribution of stakes, this time the *Pareto* distribution:
6/ As you see above, there are now a few laggards. The #Avalanche consensus does *not* always manage to synchronize all honest participants. The reason is due to the "Law of Large numbers" (LLN); let me explain:
7/ In probability theory LLN is a theorem that describes the result of performing the same experiment a large number of times.
8/ According to the law of LLN, the average of the results obtained from a large number of trials should be close to the expected value and will tend to become closer to the expected value as more trials are performed.
9/ This feature of LLN is what #Avalanche exploits and what makes it also so fast. However, LLN has its *limitations* as we saw in the *Pareto* distribution:
10/ The average of the results obtained from a large number of trials may fail to converge in some cases. For instance, the average of n results taken from the Cauchy distribution or some Pareto distributions will *not* converge as n becomes larger; the reason is *heavy tails*.
11/ The Cauchy distribution and the Pareto distribution represent two cases: the Cauchy distribution does *not* have an expectation, whereas the expectation of the Pareto distribution (α<1) is *infinite*.
12/ So LLN does not work with the Cauchy and Pareto stake distributions. This is the reason, why #Avalanche cannot guarantee liveness: In such cases it may just take a little longer.. but we saw that we can easily fix this by voting with mapped ("virtual") stakes:
13/ Here we have taken the 3/4th power of the Pareto distribution, i.e. (scaled) double square-root of cubed stakes:
14/ Above, liveness has clearly improved and in 1000 simulations over 10'000 validator nodes *every* single simulation managed to synchronize all honest nodes.
15/ Not only did liveness improve, but decentralization has also improved because by taking the 3/4th power of Pareto distributed stakes, we took away some (virtual) voting power from the large validators and gave it to the smaller ones.
16/ So, as shown it is possible so salvage #Avalanche consensus even when the stakes are horribly distributed. But there is still a problem:
17/ The #Avalanche *consensus* mechanism is fantastic to synchronize honest nodes. But it assumes that the majority of the stake-weighted nodes is honest. If however the stake-weighted majority is controlled by a single entity - like e.g. the #Avalanche foundation..
18/ ..then we have a problem: A completely valid transaction could be blocked (i.e. censored) by the #Avalanche foundation nodes. The *consensus* mechanism is *very* efficient, but it is _also_ efficient for good *or* bad decisions!
19/ And since the current distribution of \$AVAX is so blatantly distributed with a GINI of 85.6%, it *consensus* mechanism has *always* a bias in favor of the larger stake holders validating transactions:
20/ The reason the #Avalanche foundation is holding so much \$AVAX is because 360M \$AVAX have been *PRE-MINED* and very little has actually been sold to "outsiders", while the "insiders" have kept most of the funds.
21/ This means that the *Proof-of-Stake* _sybil_ mechanism protects the network from malicious actors, but it also *centralizes* the network. And since so far, *all* validators rewards are proportional, this *gargantuan* stake imbalance will remain in place.
22/ Now there are ways to *fix* this stake imbalance as well; check out these blog posts of mine at calaganne.blogspot.com/2021/05/avalan… and at calaganne.blogspot.com/2021/05/avalan… - none of which have been implemented yet: it boils down to *taxing* the rich & rewarding the poor, like in any real economy.
23/ But, despite the #Avalanche foundation having really messed-up their PoS sybil protection, the technology (or at least vision) is so *good* we can still salvage the platform with the help of *subnets*! Let me explain:
24/ So, summarized: #Avalanche the consensus mechanism *works* very efficiently to synchronize validators, but due to the bias introduced by the *PoS* protection against external malicious actors, it leads to internal centralization by the foundation.
25/ But *Proof-of-Stake* is *not* the only sybil protection in town! We still can combine other protection mechanism with #Avalanche. One, that #BTC has proven (over a decade) to be robust enough is *Proof-of-Work*!
26/ Actually, even the *original* #Avalanche paper makes this point: "The #Avalanche *consensus* mechanism can be combined with any *sybil* protection", like PoS, PoW, or even weird stuff like Proof-of-Spacetime.
27/ Now, that means we could start a new #Avalanche subnet and use the same *Proof-of-Work* sybil protection like e.g. #Monero does with their Cryptonight algorithm. How? Let me explain:
28/ The Cryptonight PoW is (according to #Monero) centralization resistant! So, a validator could mine in the background *voting rights*, and when a #Avalanche polling request comes in, exercise those rights getting in exchange some tokens, let's call them *POWER* coins.
29/ These *POWER* tokens would be really valuable, because you can only create them via PoW and then help to secure the subnet. Thanks to #Avalanche you still would have sub-second finality.
30/ I actually image a "Create Subnet" button in the #Avalanche wallet, where I could then select some parameters including (via a drop-down menu) from various *sybil* protection mechanism and then power up my truly *decentralized* #Avalanche sub-network!

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!

# More from @notexeditor

27 May
State of #Avalanche: a) number of validators at 982, 😎 b) TVL at 258.7M \$AVAX, 🤑 c) GINI at 85.6%:
WTF: TVL is *down* by 40M \$AVAX in the last 7 days? I guess somebody is pumping up validators numbers, and trying to reduce the *apparent* GINI I'm measuring.. 🤣 Note that *real* GINI is by definition _worse_, due to the top validators belonging to the foundation..
1/ Why the socio-economic structure of #Avalanche and #Turkey are similar? Let me explain: