tag:blogger.com,1999:blog-8436330762136344379.post2352822870524046998..comments2024-03-29T03:33:17.674-04:00Comments on Metadata: Snowflake to Avalanche: A Novel Metastable Consensus Protocol Family for CryptocurrenciesMurathttp://www.blogger.com/profile/07842046940394980130noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-8436330762136344379.post-30402118950671329382019-02-17T10:36:47.748-05:002019-02-17T10:36:47.748-05:00The problem with avalanche is that it assumes a hi...The problem with avalanche is that it assumes a high level of node participation (the "membership" must not change too much, section 3.7). In other words, there's no protection against network splits. The authors said they would address this in a later paper, but that's the hard part. It solves C and A but assumes there is no P. Ava and BCH said partitions are not part of the real world, but if they are not a big issue, POW did not need inventing. POW's magic is proving the chain history had the least sum of partitions with only 1 member per election needing to communicate that he won and everyone immediately agreeing without communication. No membership list is needed because it does not prove there was no partition, it only proves it had the route of least partitions, assuming a 51% attack has not occurred. <br /><br />So if there is a network split that coincides with conflicting spends on each side, the network is permanently forked. There's no mechanism to tell nodes which fork should is correct unless it defaults back to POW, but if it defaults back to POW, there is no protection against a normal double spend (I have BCH in mind). BCH miners can opt to include unseen txns or exclude Avalanche-approved txns, but this does not resolve the issue.<br /><br />I have a scheme to enable Avalanche (and can be modified for all POS & POW schemes) to measure participation in order to prove there was not a network split. Stakers must register their time-locked stake with a txn. For each txn that comes up for a vote, they hash the txn plus their registration txn a small number of times that is proportional to their stake. They transmit their lowest target result if it is one of the 30-lowest targets seen. By averaging the 30 lowest targets seen, every node knows independently (even if they get a different set of 30) the degree to which the stakers were participating, giving the node the go-ahead to approve the txn vote and requiring miners not to select a different txn. An attack would require a 51% majority on stake and POW. But if only 70% of registered stakers are shown to be participating, a 35% staker could attack. There are a lot of complexities in trying to get this to work.Zawyhttps://www.blogger.com/profile/03727573717462028189noreply@blogger.comtag:blogger.com,1999:blog-8436330762136344379.post-81533744956655636452018-06-26T10:51:58.283-04:002018-06-26T10:51:58.283-04:00Would a heterogenous compute environment (e.g. IoT...Would a heterogenous compute environment (e.g. IoT devices vs. m5.large EC2 instances) provide any challenges? Would a light-weight compute instance have a decreased chance of getting its transaction accepted, for instance?Duane Johnsonhttps://www.blogger.com/profile/05018718028789144349noreply@blogger.comtag:blogger.com,1999:blog-8436330762136344379.post-90803510414736044422018-06-05T14:18:45.405-04:002018-06-05T14:18:45.405-04:00Nice review of the paper. Would love to discuss in...Nice review of the paper. Would love to discuss in-person at Blockchain Buffalo meetups. http://www.blockchainbuffalo.com/synacenoreply@blogger.com