Pay-for-use pricing?
Just curious, I think this would be pretty cool.
Imagine providers averaging your CPU usage over the course of a month, RAM, disk, b/w and then charging based on usage only.
If you're idling a VPS, cost will be much lower than if you're transcoding on the CPU (and from a provider's side, it makes sense bc you charge more for people who use more resources)
What do you guys think?
Comments
I think you will get a lot of "Oh, that much, sorry can't pay".
I'd love this usage based pricing model.
CPU, network I/O, and IP leasing are easily measurable in all cases.
RAM usage only makes sense for containers (LXC and OpenVZ).
For full virtualization (KVM), it doesn't make sense because the guest OS would take up all the available RAM as filesystem cache.
Nevertheless, I would like to have an API that seamlessly resizes my RAM (with or without reboot), because some of my apps require more RAM during installation and upgrades (NPM, pip, Go Compiler, etc), but less RAM in daily operation.
It is also difficult to measure disk storage in full virtualization, if the guest OS is accessing it as a block device and the hypervisor doesn't know which blocks are empty.
However, I/O is easily measurable, and the amount of writes is a useful metric in billing because it affects NVMe storage lifespan and hence provider costs.
A bit more than list, start, stop?
A bit less than SLURM? 
Does there exist a simple, open source implementation that @yoursunny might like?
Tom. 穆坦然. Not Oles. Happy New York City guy visiting Mexico! How is your 文言文?
The MetalVPS.com website runs very speedily on MicroLXC.net! Thanks to @Neoon!
You mean like this?
https://aws.amazon.com/fargate/pricing/
This a is container not a VM but it’s the same concept.
Sounds a bit like what the whole "Jelastic" approach is about.
Would very much be into this kind of model.
Was thinking the same thing. Currently using cloudjiffy by @leapswitch which requires payment per cloudlet which is 128M ram or 400Mhz CPU (whichever bigger)
IMO, pricing like this only works with Enterprise or people who want to save money / use a tiny bit.
Just out of curiosity: what are you deploying/running there?
I only played around with the Jelastic service provided by ArubaCloud, but even though I liked the idea/system in general, I couldn't really find a use for my (non-enterprise) needs...
GCP and Azure also have something like this:
https://cloud.google.com/run/
https://azure.microsoft.com/en-us/services/container-instances/
Contribute your idling VPS/dedi (link), Android (link) or iOS (link) devices to medical research
Seems like it's about $10/month for .25 CPU and 512MB RAM
Personally I like predictable pricing so I'm not a fan of pricing models like this. I want to know exactly how much my bill will be every month. For CPU usage, I like when the host is straightforward about the limits (eg. BuyVM's lowest plan clearly says "Average usage of around 15% of a physical CPU"), and then customers that use more than that can just upgrade to the next plan. I like hosts like BinaryLane where you can add more CPU cores without having to upgrade everything else (RAM, disk space, etc), although some of their nodes are very unbalanced as a result.
This is already somewhat common with bandwidth (95th percentile billing). https://en.wikipedia.org/wiki/Burstable_billing
Daniel15 | https://d.sb/. List of all my VPSes: https://d.sb/servers
dnstools.ws - DNS lookups, pings, and traceroutes from 30 locations worldwide.
Hello, I'm a co-founder of Jelastic PaaS. Our team has invented the real "pay-per-use" billing model in the cloud computing world and tested this approach during the last 10 years. Customers who understand the benefits and how it works love it a lot.
It suits not only for enterprise applications but for any workload that is above cheap standalone (e.g. $5 VPS). One of the the key advantages of such an approach is that the real "pay-per-use" solves the right sizing problem.
Regarding AWS Fargate:
AWS does not offer a real "pay-per-use" model. They say "you only pay for the resources that you use" on the various related web and landing pages but it is a misleading message as in reality you always pay for allocated (not used) resources.
Same for AWS and GCP - they charge end customers for the allocated resources, or in other words for the limits, not for the actually used resources. So, they confuse its customers.
Any provider that sells containers using the old school fixed price or even "pay-as-you-go" model shares allocated but not consumed resources with other containers on the host machine as they are not reserved by default. We believe that it is not fair to charge users for the resources that are re-sold to others.
If you have a small number of instances then predictable or so-called fixed pricing works fine. But as soon as your projects start growing, getting load spikes or variable load, scaling vertically or horizontally, fixed pricing does not work well anymore.
There is interesting feedback from the CEO of a digital agency that runs and manages multiple customer projects using “pay-per-use” pricing .
Also at this session we discuss the basic points and advantages of “pay-per-use” billing model and vertical scaling.
Please let me know if you have additional questions or concerns related to this topic.
The low-end buying ethos is about getting as much resource for as little as possible. As such, a ‘pay-for-use’ model isn’t in keeping with the mindset of the buyer.
I liked the hourly model for cloud services. Wouldn't mind it as an option.
It does work better though in contexts where there is a lot of stop/start automation & autoscaling though
Scaleway already has hourly, no?
I do not see the conflict between "pay-per-use" and getting cheap resources. Can you rephrase?
Start/stop automation and hourly pricing model are applicable to pay-per-use as well. However the real pay-per-use is more about vertical scaling and granular measurement of the used resources.
Pay per use conflicts with the tenet of idling your resources
No. You’re doomed to forever wonder.
None, tbh. I’m fine with VirMach or SSDNodes (yes, them) most of the times, and when I need some stability, I go with Vultr or Lightsail
Do the "fixed" and/or "pay-as-you-go" models work better for idling resources? Can you clarify your statement?
For most LE stuff, just paying for resources and getting them to you oversold basically amounts to the same, mainly since the LE market already runs on such low budgets.
Managing the node when resource usage for multiple users scales up and down sounds like a nightmare that a provider wants good payment for, and as a result, pay as you go might turn out more expensive than pay upfront.
Hey teamacc. You're a dick. (c) Jon Biloh, 2020.
For sure "pay-per-use" requires an advanced automation and orchestration than actually give cloud hosting service providers a competitive advantage compared to the other providers who sell resources in the old way based on the fixed model with overselling. And yes, the first service providers (the pioneers) usually increase the base price because they offer something new and unique on the market, however as soon as this model gets widely adopted the price goes down as usual.