Can you advice me in cheap good quality VPS services? Maybe we can build a list here for reference?
The cheap services have a lot of variance in their quality at least as it relates to presearch reliability scores and PRE earned. Some nodes can perform well while others consistently underperform even within the same provider and quality is not clearly indicative of higher rewards.
There are several very good vps, oracle, ethernetservers, hostvds, ionos, etc.
This is what I am talking about with quality. Yes you can get cheap VPSs for nodes but cheap is not great for the network. Cheap plans as shown from hostVDS often limit bandwidth (50 Mbps and even 200 Mbps is junk and slower than both of my personal home internet connections [500 Mbps and 1Gbps fiber]). The users of Presearch have benefited from Community members providing higher quality and better bandwidth nodes currently at a loss. Those nodes WILL GO AWAY unless nodes are compensated for the quality given to the network, and when they go away search results will likely decline in ms speed to fetch and return to the user or may see increased failed results. Quality VPS nodes are $5-10 / mo minimum as seen with pricing at some of the better more reliable providers. some offer flex pooling and even free vps capacity up to a limit but quality is not cheap and current monthly returns at .03 PRE does not cover the costs.
I recommended the Team create metrics to track the number of failed attempts and ms response time of every search query. If the project is serious about providing snappy quality results they should be tracking these simple metrics and work towards decreasing both not increasing them. If they are tracking these I think this current node decision that will start in September will show a trend of lower quality in both metrics. We need to compete and provide as close to or better than what people are currently used to getting if they come over and they are waiting a second or 2 for search results or for some reason a fetch fails this is catastrophic for the project and network.
2 Likes
The above doesn’t account for storage or GPUs that will likely be required for other future node roles or AI. Some of my nodes are providing 1TB of SSD capacity on a single node. How far do you think the project will get with indexing, storage, or AI with a bunch of shared 1 core CPUs, 1 GB mem, 10-25GB capacity VPSs of which half of the capacity is already used up by Presearch docker and the ubuntu operating system? The answer is not very far. Quality in memory, CPU, SSD capacity, GPU (AI), Bandwidth, IPs all costs substantial resources and are needed for the success for the network.
I am not advocating that rewards shouldn’t be brought in-line with income. I am simply saying if you don’t adjust how you cut costs and don’t change the current model there will never be incentive for quality. All nodes will be run for economics if quality and other factors important to the networks reputation and resilience are not incentivized.
1 Like
This post is filled with poor assumptions. Half of my nodes are Racknerd (“cheap” option by your definition as it is only 2 dollars a month). They all have zero failures, reliability scores averaging 85, and are completely underutilized.
We do not have a node quality problem. We do not have a serving capacity problem. We have an expense/income ratio problem and here you are arguing for why presearch needs to spend more money for nodes that are underutilized. You’re even trying to argue that we should pay for storage and GPU which would be for projects not even on the road map at this point.
We need to walk before we can run. Paying for higher quality, higher capacity servers when we already have too much aggregate capacity makes little sense. Paying for hardware that the project cannot currently use makes absolutely no sense.
3 Likes
I’m not trying to avoid cutting costs to an equilibrium with income. All I am saying is there are other important considerations/factors that should be prioritized and taken into account with the node rewards. So the way you cut rewards matters and the way you move forward into the future also matters. The amount of nodes using racknerd probably exceeds 10% of all nodes this creates a potential centralization issue and vector of attack or weakness to the network. The flat cut forces centralization into fewer low cost VPS providers so not only may service quality be reduced but also the only part of this project that makes it decentralized will become more centralized as well.
You don’t have to agree with me I am sharing my concerns with the community and team because someone has to do it.
I mentioned storage and GPUs as future cost considerations not that they are required today for search queries. But since the team has both AI and Storage nodes on the road map neither of those are too far off from development and early
testing.
1 Like
I understand your concerns, and there is of course some truth to what you’re saying. However, these are not Presearch’s most important problems by a long shot. Nobody is going to abandon PRE because their search result takes a few milliseconds longer to return.
If Presearch becomes much more successful, some of the items you have raised will start to matter. The good news is the reward algo / economics can be updated to drive such changes if and when that need presents itself.
2 Likes
You can also use Ethernet Servers for $1.95 per month (paid yearly). Hopefully this is allowed, here is my referral link: Lightning-Fast SSD Web Hosting | Ethernet Servers Ltd
2 Likes
Thank you for all great suggestions!
You can get a cheap Ryzen gddr5 from Racknerd for like $35/year or a KVM gig or so for $15-20 a year. Really helped me cut costs
1 Like