If the geolocation of dYdX validators was very globally decentralized as well as the institutional users of dYdX you could then talk about randomly choosing 30 validators to eliminate, which could be the bottom 30. However, and this is key, the majority of institutional users of dYdX are near the Tokyo area as well as a good number of validators. In this situation to improve performance you CAN’T just randomly pick 30 validators to eliminate like the bottom 30, you must identify which validators are based far from Tokyo which in dYdX are around 30-40% of validators and eliminate those selectively if what we want really is to improve performance
The thing is, the author of the report @nethermind used an arbitrary assumed infra cost of $1k/month to justify that so many validators are running at a loss but this assumption is totally wrong and shows a very poor understanding of validator costs. There are three main types of servers: bare metal which you can rent at data centers, colocate in data center or run in your own data center, then you have Dedicate virtual servers which are virtual servers but not shared but still less performance than baremetal, and then Virtual private servers which are virtual servers shared and have the worse performance. For dYdX specifically since many are based in Tokyo it is important to highlight that bandwidth in Japan is very expensive, now to the analysis.
Of course if you go with AWS, GCP and similar you will pay insane high costs for the most simple and lowest specs servers. Also, given the block time of dYdX let’s assume virtual private servers or virtual dedicate servers wouldn’t be the best for performance but these would be the following options:
Without giving names, you can find providers of VPS and VDS for monthly costs much lower than the suggested $1k, in fact you can find very powerful VDS for less than $100. Moreover, you can find top bare metal servers in great data centers in Japan for around $100-300. If we add unlimited bandwidth in Japan some providers offer this for around $300, so you could have a top bare metal server with unlimited bandwidth in Japan for less than $500. But still a top virtual dedicated server with very high bandwidth and specs could be found around $100-200 in total, very far from the suggested $1000. Of course if some go to AWS, GCP or the most expensive providers and pay $3000-$5000 for the same services that’s their choice. But don’t imply that every dYdX chain validators has $1K infra costs per month. We use a top provider with bare metal server, unlimited bandwidth in Japan, we are consistently in the top 10 dYdX validators by best uptime and lowest MEV and we don’t pay monthly anywhere close to $1k for the infra costs, it is more like half of that to get the best possible performance, but for other validators with lower specs or not unlimited bandwidth they could easily have just around $200-300 monthly costs. So your whole analysis of validator profitability and suggestion to remove 30 validators based on that flawed analysis is just totally incorrect and inaccurate.
From the report: ‘The profitability calculation below assumes similar fixed infrastructure costs of $1,000/month and the staking APR net of emissions (inflation) has been considered. We acknowledge that infrastructure costs could be higher or lower, but use $1,000/month for simplicity.’
You use $1000 ‘for simplicity’, great argument. You didn’t conduct any research or survey to estimate how accurate is this $1k. The costs would be much higher yes like $5k or more and the whole dYdX active set would be unprofitable if all go to the most expensive providers possible. Companies don’t open factories in Switzerland with the higest costs and labor costs in the world, but they choose low cost locations. Similarly, the majority of validators don’t go to the most expensive providers for the same services, maybe Polychain or similar validators yes, but the great majority no. Which means most dYdX validators have costs much lower than $1k, not $1k or higher than $1k, therefore your profitability analysis is totally wrong and if it hadn’t been discussed and corrected your flawed analysis could have caused great damage to dYdX by eliminating many top contributing entities based on flawed assumptions