Feature proposal - Provider profiles

At the moment, I choose my provider mainly according to the bid price. But since there are often several bids, it is difficult to decide on one. Often I just randomly pick one, and I think others will feel the same way.

Especially when it comes to hosting there are more parameters to consider than the price. It would be practical to have profiles of the providers in order to be able to better choose the provider.
It could include following points:

  • Hardware (Intel, AMD, processor type, etc.)
  • uplink (100 Mbit/s, 1 Gbit/s, …)
  • Hosting location (data center or the own garage)
  • Maybe a small custom description why someone should choose the provider
  • Contact/Communication thread (on case of outages)
  • Many data centers nowadays operate with renewable energy. Maybe that could be added too

Those would be my first thoughts. Of course providing those informations should be optional.

7 Likes

Would be cool indeed to have more information on the providers!

3 Likes

Hi,
these are surely useful features. We had similar questions, and you posted here, thanks for that.
I would like to suggest more in this regarding.
There can be a grouping/labeling of server-class based on the server hardware like

  • blue, green, black
  • seed, plant, tree
  • or something like starter etc.
    With this grouping, the providers can be filtered based on class, and then the user can select from the features suggested by you.

Like this is just our thought, maybe someone has a better suggestion.
Thanks.

1 Like

I like the idea of grouping it. It also would maybe create the possibility for higher ranked provider to place higher bids. Don’t get me wrong, I don’t want higher prices. But I believe a provider with modern hardware, a good uptime and 1 Gbit/s uplink should be allowed to charge more than a new provider with 100Mbit/s uplink and some older hardware.

Hi,
Yeah that is a possibility but still it is upto providers to charge and we to select a friendly bid for us.

But grouping will help people for specific needs, rather looking for all bids.

A case be like more compute less ram, or any specific configuration.

The grouping will help non tech people more in terms to this.

Hey, I think that as akashlytics evolves, it would be cool to offer providers the opportunity to register and award labels in order to stand out. In addition to having just a simple list of different providers.

1 Like

What would you do with knowing the Hardware specs? What are you trying to determine there?

With the uplink data? What’s the purpose of that? Bandwidth is shared so knowing the bandwidth doesn’t necessarily tell you anything about available or performance, no?

A provider would also want to know about Tenant profile. They will have a reputation to protect and they may not want certain types of sites or workloads to run on their infrastructure (will create a separate post for this if it’s worthwhile to discuss).

I draw a bit of a comparison to centralized hosting providers here. My first thought would be the CPU. For me as a customer it matters if I am running on a new powerful CPU or rather an old one. For instance this old one vs this newer one, just as an example.

After that, the download/upload speed would also be important for me. Is this rather the connection of a computer at home (with 100Mbit/s) or a connection like in a data center with 1 Gbit/s transfer rate?

Furthermore specs regarding the disk might be interesting too, HDD or SSD?
The next point might also be if the provider is using renewable energy. In today world where sustainability is very important this shouldn’t be underestimated.

Now I don’t know enough about Kubernetes to exactly tell if that’s useful what I am requesting here. But anyways some more information about the provider would be good in my opinion.

I also thought if it wouldn’t be possible to create Benchmark Containers to be able to rank providers regarding their performance. Right now I’m literally choosing my provider by randomly picking one

True, right now, it’s primarily price that’s the driver.

What’s a different decision you would make if you knew old vs new or 1G vs 100GB? Would you pay more for one with better specs?

These are all good criteria. The bandwidth is tricky to get from Kubernetes I think. The rest is not trivial. Kubernetes is a cluster which can contain many different machines of different specs. K8s schedules the workload to where there is space available based on what you want for memory, cpu, disk, etc.

One way to have a managed monitoring service that monitors the workload performance and based on that you can switch providers, pay less, pay more, etc.

One of the benefits of being “serverless” is you don’t need to worry about provisioning different spec machines. You just specify what your workload needs.

The things like latency (ping in ms) perhaps would make sense based on the geo for example?

Sustainability for sure is a criteria you would decide on. That makes sense. A second criteria is power consumption (aggregate) because the provider may not be dedicated to Akash. Would that be decision criteria to use or not use that provider? - I don’t think so. Security related things are a legit criteria to make a decision of choosing I think.

1 Like

Depending on my use case, I would be willing to pay more for a better server.

Now let’s say (this is a bit overkill) I want to run an Ethereum node. Then it is already important for me to have an SSD in any case. I would also prefer 1 Gbit, as well as a somewhat higher-performance CPU. If I’m only looking for a web hosting for a blog, then I don’t really care.

Bandwidth and performance could also play a role for online video conferencing/game servers and similar. Especially if the target audience is a bit bigger.

Regarding latency the geo information is also quite important, yeah. But isn’t that already implemented?

Personally, I don’t know if the actual power consumption would be an important criteria for me. I never really thought about that topic so I don’t think it would be that important. But that’s just my personal opinion.

Can anyone actually become a provider or do they have to be approved by the team?

How would you validate each provider profile? I like this idea but not sure having this info on-chain is the best approach. For instance, things like provider up-time, latency etc. might be best independently verified.

1 Like

That’s the purpose of audited attributes :slight_smile:

@crtl-Felix - Today we are the auditor. In the future, there will be other auditors so that becomes a selection criteria as well.

Wrt to geo -yes, today we have region. There is a standard for datacenter Tiers. 1, 2, 3, 4 based on reliability. The auditor can vouch for that so that’s one way you can filter among hundreds or 1000’s of providers.

For production workloads, only show me bids from Tier 4 providers.

For test workloads, show me bids from any provider or only Tier 2 for example.

When we support persistent storage, the storage class: SSD, will allow for setting SSD vs HDD etc. WIP and subject to change.

Good point about audited attributes. I guess I was thinking of something more real-time. I.e. was this provider certified once and then had horrible downtime, bad upstream router problems, etc. But thanks for explaining. This sounds like the audit process could evolve to support something like this.

Was about to post another feature request similar but I’ll piggy back here (let me know if this is veering to off topic). I’d like to suggest that certain providers may also want to include “capability” profiles such as a less restrictive container run-time.

Currently any sort of non-standard application such as VPN or network monitoring will likely have challenges running withing gvisor.

Probably a separate FR

Security is an area for exposing capabilities where a tenant likely will want to filter providers on, based on the type of workload they want to run at the time.

Are you asking that you would want to know if a Provider supports gVisor?

Will post all this in another thread, but yes for starters listing the container runtime would be nice. But in a larger sense if a provider is comfortable allowing additional (capabilities, they could charge more and many would pay for that extra service.

1 Like