New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
What does fair share CPU mean in the big 26
Hey guys
Quick question on what is actually considered "fair share" of CPU on a shared VPS.
Obviously it depends on the provider policy, but let’s say they are a budget provider and they just have some generic “no 24/7 use, no heavy use, …” terms.
If you need to run a continuous background workload, what % of your allocated compute do you expect to use for long periods without getting throttled?
I've seen people in old threads say 50% average load is fine. Even there is this thread from 2016: https://lowendtalk.com/discussion/73738/what-does-fair-share-cpu-usage-means. It gives me the impression they were much more lenient about it back then.
So, what’s the real rule of thumb you would stick to?
Poll
- How much usage do you think is fine?38 votes
- 0%, I only idle 24/713.16%
- 30%34.21%
- 50%36.84%
- >50%10.53%
- Other5.26%
Comments
All providers should avoid vague "Fair Use" policies without strict guarantees (especially for "fair use unlimited" bandwidth)
The fair share CPU in the big 26 is the same fair share as in the big 16 (10 years ago). Due to the pandemic and the WW3, the computing technology has not evolved much — on the contrary: there's hardware scarcity due to AI.
To try to answer the question, the best example in my mind would come from Virmach: "cannot burst CPU to 95-100% usage for more than fifteen (15) minutes and cannot average higher than 50% usage within any two (2) hour period". If you apply the same rule to other providers, I assume you'll be just fine.
To be honest, I dont like fair use policies, whether they are for CPU, bandwidth or IO. Providers should clearly state what level of continuous usage is actually allowed (like 5%, 10%, etc....).
I find statements like, "You can use 100% CPU day and night as long as other users on the node are not affected" really unhelpful. How am I supposed to know whether my usage is affecting others or not?
It would be much clearer if providers stated something like: "Bursts up to 100% are allowed, but sustained usage is limited to 5% of a CPU core on this plan." That gives customers a concrete expectation instead of a vague fair use policy that can be interpreted differently by different people !!
A server is not like a slice of a bread being cut equally for every customer. Sometimes a customer needs more CPU and 5 minutes later another customer needs more CPU. Fair share policy balances this while keeping costs low. If every customer uses their VPS to the maximum, there would be a huge performance issue.
If you want your slice and want to enjoy your whole slice all the time, then grab a VDS, or step up and grab a dedicated.
But let’s say a provider oversells CPU by 5:1. In theory, there would be no big issues if everyone used 20% compute 24/7, and they could state that clearly. In reality it wouldn't scale perfectly, but a number can still be given.
For budget hosts with shocking levels of overselling, their policies will obviously be much stricter. It would be responsible to state that dismally small number upfront instead of pretending the box can be used for anything other than idling.
It depends on your usage. You can still use an extremely oversold VPS, just not for something productive. A heavily oversold CPU has the great advantage of being extremely cheap, but the disadvantage of being very slow due to multiple customers being crowded per core. If you need some examples with regards to what you can self-host, there are plenty ideas, but you must keep in mind the load which you might need.
It usually boils down to money. You want better performance, you need to pay more. The same goes with dedicated servers: if you want a better dedicated with latest modern CPU and a maybe graphics card included, you need to pay way more. The more resources you need, the more you pay.
The purpose of fair share is to have as many customers as possible and reduce costs for customers, because the root idea is that not all customers need 100% CPU for a small website, for some personal VPN or ZNC instance, or for some productive small start-up project. Meanwhile the provider keeps a balance as to not impact customers too much (this is where overselling comes into play). If provider turns greedy, then customers will leave bad reviews, so even if you regard "fair share" as some way for providers to abuse it, it will rarely happen, because forums like this exist and the benchmarks will bury the provider in bad reviews.
fair share means you are not allowed to use more cpu than users on average use. When you use more than 20% of cpu over long periods you will be kicked off. If 20% is the average use. Often the average use is less than 5%.
I barely hit 5% CPU so I voted for 50+
The main issue is failure of setting expectations, failure to disclose limits and failure to disclose consequences of breaching limits by providers.
Perhaps providers are afraid of looking bad when honestly advertising what they are selling.
Provider can offer a Prometheus endpoint so you can query realtime host CPU usage.
We created 48 VMs, 2-core each, on a 6-core physical machine.
It’s free VMs for computer networking class.
First year students cannot tell their VMs are massively oversold.