Given that this is in research - I’ve been thinking about how this ties into V0, V1, and our goal to full decentralization.
While this is a great thing and keeps us up to par with centralized providers, we may end up where the network relies on a handful of centralized providers for edge networking.
Allow me to elaborate - right now, there are very few competitive “edge” centralized providers today - Cloudfront, cloudflare, route 53, etc that retail customers can leverage. There are a couple of other CDN providers up there such as Fastify, akami, etc as well to call out for larger clients. Either way, if the network starts consolidating into these couple of providers, most of of POKT traffic will terminate at their PoP’s and potentially traverse their backbone only.
It isn’t a surprise and pretty common for networks to traverse the same IXP’s, but as the cherry picker continues to reward strictly off latency, we may see a threat where all providers only use 1-2 centralized providers for their networking in order to be competitive. Without fear mongering, I have worked in the cloud space long enough to see that even the best of the best can have their fair share of outages. While POKT wouldn’t be the only service impacted (given how many websites are powered behind large CDN providers) - it does conflict with our long term mission. This is definitely worthwhile calling out as we continue to navigate the space and how we design the next generation protocol / cherry picker updates.
This is an excellent point. It also raises the question: is low latency more important than decentralization? If so, the network will naturally centralize to whatever configuration provides the lowest latency because that’s what pays the highest rewards.
The million dollar question, and I would not put it behind us to solve for both in the long term future.
Thinking about my post earlier post- this just occured to me that this type of deployment could tie into the recent events of the sanctions of Tornado Cash and centralization in general. I am no lawyer - but it seems plausible that if data flows in between Cloudflare PoP’s and AWS PoPs, isn’t there an existential threat that they could start banning users who do not respect these sanctions? This definitely conflicts with our current incentive models where you are rewarded for low latency and encouraged to use edge servers, could we see a threat where the entire network blocks these calls long term with the current model + multi - region deployments?
Although I have not seen it written exactly like this, my interpretation of Pocket’s vision is:
If that’s accurate, “unstoppable” is the priority, followed by “any blockchain”. It seems to me that low-latency and high-security are table stakes. Performance and security are critically important, without a doubt; but they are not strategic differentiators.
If the priority is in fact “unstoppable”, then the technology and incentives should always align to that end. If low latency is the primary driver, in all likelihood, this will make achieving “unstoppable” exceedingly more challenging.
Disruptive technologies need to prioritize the “disruptive” feature first and foremost. For example, NVMe drives are disrupting standard SSD drives because they are way faster - not because they can compete on capacity or storage costs - yet. Of course capacity and storage costs are important considerations. But it’s the speed of the NVMe drives that makes them attractive when compared to standard SSD drives. In the same way, if “unstoppable” is the unique value proposition for Pocket Network, then “unstoppable” needs to always be the priority.
To that end, we should always be asking:
Does this [incentive | technology | whatever] help make the network more unstoppable?
I think that the objective of being an unstoppable and decentralized network does not collide with being a fast provider.
It is true that many node runners are using AWS for hosting their nodes in the same rack as the portals (if they can afford it), it is also true that many are using CloudFlare or similar services to increase their speed (reduce portal to node latency). This is indeed against the decentralization objective of the network, and this is a consequence that right now the latency is the only metric that is being used to distribute rewards. However I think that this is not very alarming. for the following reasons:
The current Cherry Picker settings make every node under 150 ms of Weighted Success Latency equally eligible:
You don’t need to be in the same data-center as the portal to be a high QoS node.
The mesh nodes make more affordable to have chains servers around the world (more utilization by all your nodes). Smart routing services become less critical. Node runners can focus on deploying chains and keep their Pocket nodes in cheaper (or permissive) locations, using public routing.
Our ecosystem is versatile, heterogeneous and over-provisioned. We are much more likely to resist disruptions (or sanctions) than any other service out there. Our only problem (currently) are the centralized gateways.
I think that the main force pushing the centralization roots in a very dark place that I don’t like to enter, the economics. The Pocket Network is increasing the pressure on node runners mostly due to its inflation control mechanisms.
The POKT emission is firmly controlled, despite the growth in relays, the pie becomes always smaller. Small node runners have an unfair fight, as the cost of running a chain is too large if you don’t have enough nodes to exploit it. Big providers have the power of the scale. The question is, how much pressure can these little guys support before they disappear? Sadly I cannot tell.
You might think that the GeoMesh added additional pressure to the little guy, but as we said in our initial post, some large node runners already had this capability before us. We only tried to even the fight. Being secretive always favors centralization, the only thing that we can do to help the small node runners is to make things easy and performant for them.
I’m not saying the two “collide”, I’m simply saying that I don’t believe low-latency is what will diffiernante Pocket from the competition. Further, it’s my opinion that while the rewards are largely driven by performance and not decentralization, the network will tend to get more centralized - not less.
This is not a criticism of GeoMesh or anyone wanting to optimize to improve rewards. We plan to implement GeoMesh at Dabble. That said, it is simply a fact that on a globally distributed network, some nodes will always have a distinct latency advantage over others because of their physical location. So, if incentives are largely tied to latency, nodes will most likely cluster towards locations that provide the lowest latency.
In addition, if technology solutions like GeoMesh further incentivize this clustering by lowering costs, that will accelerate the clustering effect in all likelihood. This will be quite easy to validate over time but there is strong evidence to support this trend already.
I agree that the root of the problem is economics. I also sincerely appreciate the fact that POKTscan is trying to help level the playing field. However, I still expect that GeoMesh will likely accelerate node clustering.