DigitalOcean for dedicated servers? Good or bad idea?
Hello IHers!
I wanted to run this by you to gauge interest. I'm working on an IaaS like DigitalOcean but for dedicated servers instead of VMs. So developers will have the ability to slice one or more dedicated servers into however many VMs using an intuitive UI (similar to DOs).
I feel that devs get discouraged from dedis in general because of the ops involved getting it to do what they want. But if this service were to simplify that part of dedis to the point that they don't even have to think about it, would that help?
Please take the time to fill out this 6-question survey. It will help tremendously!
https://docs.google.com/forms/d/e/1FAIpQLScJxX6CF5vyJovWeBLhrqHQ9NDo3kycTJMbwz4QUwDGT86BkQ/viewform
Update: I got surveys AND comments here! Amazing! Thank you and please keep em coming
Thanks
I think a dedicated server would be overkill for almost all devs. Small-scale devs would have no use for this -- only enterprise who require a ton of computing power, but then they would likely need storage arrays that can keep up with the operations/second that compute blade is doing...seems like you'd need a full-fledged datacenter if you were going to offer dedicated blades.
Thanks! A dedicated server with 8 cores, 16GB RAM, 500GB SSD though powerful wouldn't be overkill for devs who need to spin 10+ VMs (are devs who need this many VMs uncommon?). I guess the question is what price point makes such a server attractive to a dev for consolidation vs just getting 10+ DOs.
On the need for storage arrays, DO started with locally mounted storage volumes. They did that for many years and still do that. Nothing fancy. I know they have block storage as an option now too.
I guess the question is, with such a small physical server, what benefit do you offer the dev? Why would they choose that physical server over a virtualized one that can be grown beyond the hardware specs of your physical blade? I think that's the value question that needs to be answered.
Some use cases I can think of:
A. spin one or many small-RAM droplets (ex: 64MB). I don't think you can spin smaller than 1GB droplets in DO. Docker comes to mind.
B. Have many customers but need to segregate their deployments. Can save big time by being able to slice a dedi into a hundred VMs
C. The need to spin a VMs and work on it for a while without billing pressure. Costs $0 to spin a new VM
I can think of others but I guess it's the market that decides
C would be a fallacy though wouldn’t it? It wouldn’t cost me to spin up a VM because I’ve already paid up front, so why wouldn’t I save money and just spin the VM up when I need to use it?
The only real benefit is a consistent monthly bill that is likely to be higher than cloud provisioning VMs as needed.
It's definitely a fallacy in your described case above. But I was thinking more along the lines of a dev who's already spun 10 live VMs and is already getting more than their money's worth. So spinning additional VMs is like free gravy at this point.
I work for an ISP, and I look after about 400-500 physical ESX (and similar) servers across 3x data centres, along with tons more networking and related kit. I also use DO/Vultr/AWS for personal stuff.
I really don't get why anyone would want this. The world is moving towards the cloud, and this seems like a step backwards. If you've got the money for racks in DCs, you can run your own servers and have your team manage it for you. If you're a startup or similar, why have the hassle of managing a physical server when you make use of on-demand VMs, serverless functions and so on for pennies? Paying for a dedicated server means paying for 24/7 power, cooling, and dealing with the physical failures you'll have running a single server in a DC somewhere.
DO/Vultr/AWS are responsible for my VMs being online, I don't care how any of the hardware works underneath, as long as it works. In your scenario, if there's a HW failure (and DC hardware fails often, I know as I spend a lot of time in London with Dell engineers...) then I'm offline. If hardware fails on one of my cloud providers, it's dealt with routinely with little-to-no downtime. Not only that, but if I suddenly want to move VMs to a different region of the planet, it's easy. Floating IPs? No problem. Built-in backups, snapshots, private networking? Done. IPv4/IPv6? You got it.
Sorry to be a downer, I think this market existed over a decade ago, but now the trend is VMs in the cloud, containers (GKE and so on), serverless functions and more - dedicated servers are disposable nowadays and akin to a battery. I added 6 new servers in to an ESX cluster this week to give it a 1.5TB RAM increase, and an extra 250 or so vCPUs, then threw away 4 old servers - no real work required, just mark the servers for maintenance in VMWare and they are literally just hot swappable compute.
99.99% of people don't care about the tin, they just want 99.99999% uptime and lots of features, all for cheap.
Hey Joe
Cool idea. I think I could potentially be a customer in the distant future with my product, Codemason. Codemason helps developers deploy and host their apps.
Offering hosting as an added connivence for users has always been up on my list. I'm assuming by using something like this to slice up dedicated servers into multiple VMs, it would have a cost benefit too
Codemason looks great! I can already see how this thing can be useful to your users. Is this something you'd "inject" into your "Add Server" workflow?
Correct
The benefit is the added connivence of not needing to connect to any other cloud provider and hopefully being able to reduce costs
I think the idea is good, but a few worries that come to mind are what happens with the hardware fails (do we have a way to recover fast and easy?).
What's the cost compared to DO?
Where are the data servers located and what bandwidth is available?
I really like DO, so it'd be hard for me to switch right now, but I've switched server providers 3 times in the past 20 years so I'll never say "never" when it comes to that.
Frequent backups and point-click restores. Even DO isn't immune to hardware failures. Their SSD storage was always "local" on the server where you're hosted. If the VM goes down, you're down until you're able to restore backups.
Cost will have to be really attractive. I put a price list in the survey form but it'll boil down to what people think is reasonable for what they're getting out of it.
Hmm, at least from my perspective:
-- I don't care much about managing or provisioning a dedicated or virtual server. As much as possible, I want to focus on my application, what resources it needs, and being able to scale up/down quickly.
-- More and more of what I deploy, over the past 3 years or so, has been on containers not VMs.
Hi! In addition to other concerns, here's what would bother me: hardware dependency. Your server breaks, my 100vms are down. Lack of block storage accessible independently from said VM boxes. DNS niceties.
But here is a spin: what if this was a software service you were selling? I have a box at home and I don't wanna deal with all the setup, I just want to turn my lil box into DIY AWS dashboard without actually having to do it myself. Or I rent a pair of cheap dedis from Hetzner and I wanna splice it all up. Or I'm a web agency and rent my VM at digital ocean, but wanna slice it up between my clients, as you've said. If I could rent from you a "yum install my-iaas && run my-iaas ", it might be very interesting.
The trouble here though is that there are already things like this, so you would have to really work like a charm and have as killer feature for me to pay for the service.
Hope you find your market in any case.
The trouble
Tris, this DIY idea is great! I did think about a marketplace where you get to see and pick offers from various dedi providers but this is much better.
Can you list a couple of services/software that already offer this? I can think of VirtualBox and maybe QEMU but they only runs on local boxes.
Thanks Tris!
Well, think of it as something like hobby kube, but which fully automates every step- it only asks me for, say, domain names or oauth authorization to namecheap so that it sets things up for me.
This effectively already exists at Vultr, Packet.net
But vultr lets you create only one VM on the whole dedicated server. You can't slice and dice however you see fit IIRC.
Also, Netcup offers cloud servers with dedicated resources, which pretty much is perfect for someone looking to test things out in a dedicated environment.
So is this a competitor to something like WHMCS?
No, WHMCS is a billing system for resellers with the ability to provision a whole slew of services like shared hosting on WHM or VMs on certain platforms. I guess it does so many things but I can see how it appears to overlap on the surface.
Thanks
I think you’d struggle to find your market for this for a couple of reasons.
Firstly, those who want dedicated servers and have the budget for them will probably just want bare metal to do with as they please. It’s fairly trivial for someone who knows what they’re doing to run a Hypervisor and roll their own VMs so I don’t think there would be great demand for this. Also, see ‘managed dedicated servers’ as that’s an existing product you’d probably have a lot of overlap with.
Second issue I see with this is that a lot of the benefit of VMs/Containers is that I can spin my VMs up ‘in the cloud’ so I’m not as susceptible to hardware failure as I would be if I had everything on a single box. I get your point about spinning up micro VMs but the 1GB DO droplets are cheap enough at $5 a month (or Vultr’s $2.50/m offering) so I’d be happy enough using one of those and running containers if I wanted to isolate some very small tasks.
By all means I think you could do this and you’d have a very small niche that might attract a few customers. I just think the dedicated server market who want to eschew the very capable cloud offerings available today probably wouldn’t be your customers.
Lots of valid points here. I think that DO succeeded because they made it very easy for people to start VMs on dedicated servers - and not make them think about the underlying hardware too much. I recall people referring to DO as the cloud even though it was just servers on a dedicated server. So the idea that "those who want dedicated servers and have the budget for them will probably just want bare metal to do with as they please" is mission impossible for a lot of people (I've been running a managed server business for 11+ years).
Right now I'm seeing three interesting features based on feedback here:
Software to empower people to slice and dice their own server (at home / DIY, dedicated server at a provider of their choice, marketplace of dedis, etc)
Have it crash proof (backups, hot migrate, point n click recovery, etc)
Put users in control of their dedi by making the dedi setupand VM provisioning error proof (so as to not shoot one's own foot) and click-n-point accessible.
Actually I really doubt it. But who knows. I am used to my good webosting provider. I really appreciate https://zomro.net/ for its ssd hosting solutions on web. Before I have tested 3 hosting providers but was not satisfied. I left my last hosting company because they repeatedly made a mess of billing and servers went down a lot.
By the way, I understand surveys can be a turn off so feel free to just comment here.