sane bandwidth pricing
Eh, most dedicated hosting providers you can easily grab a dedicated server, have it be available immediately, and it is ready for you to put on an OS (and most(all?) will even give you an option of a default OS, the good ones even give you an option of no OS, pure KVM or bootp then) to do what you need, and you can shut it down and kill it with a pro-rated rate, so it is basically already there as it is, and if you have an OS image to upload to it then it is easily automated.
Very true with one big omission
Ideally on something like Juniper QFX3000-G.
they could take dedicated server pricing multiply it by 3 and still have a line of customers.
Yeah but you can still easily set that up within your own OS, have a central orchestrater or gossip protocol establish the vlan on top of the primary internet facing IP (and is similar to what I do already, though over ssh tunnels). A hardware managed VLAN would be awesome though, but it is not necessary at all. I could easily envision a good OS image you could choose that does it for you.
I think you have to sacrifice something - and it would probably be instant provisioning.
If you have a decent dedicated set-up, the couple of hours it takes to get a new server online shouldn’t be a huge deal Imo because you would probably already have a buffer and would probably be able to estimate when you’ll need more resources - especially when using a distributed system built with Elixir.
I don’t have any experience with massive web-scale apps though so this is just Imo
I worked for a company that operated its own datacenters. Although I moved before it happened, they were working on getting the couple of DCs where OpenStack ran working with hypervisor-level containerization which is pretty much quite bare metal in my book.
Throw enough money at a problem, and you can solve it. The question would be - why?
What languages were they using? Why did they want containerisation? I don’t really see the appeal of VMs myself…
Nice to have your own DCs tho!! Was it a company or app we might know?
$37 a month for 4 core 8gb ram and 80gb ssd - that’s insane - you can get a half decent dedi for that but with much better specs
There are a few mentioned in this thread:
That the thing I am not sure one has to sacrifice anything the logistics are same as AWS you have some analytics running to estimate the spare capacity you need to have on hand and as a provider you have much simpler system compared to AWS. The only high cost piece of the puzzle is the high end networking setup that you would have to overprovision to a significant degree, but it can be priced in.
Not really affiliated with them but there is scaleway, which has 3 Euros for a quite good machine. Only available in France afaik.
Thanx looks interesting for personal projects but would not cut it for work as the support does not really exist The hard ask is own VLAN that they do not provide but in the on demand format none really does.
That’s the interesting question, of course (not “what company was it” ;-)). Basically why everybody wants it - it’s arguably simpler than deploying VMs. Now, docker-on-VM-on-OpenStack is a bit too much abstraction (like docker-on-EC2-on-AWS), so it makes sense to skip the VM here and run your containers one step closer to the metal.
I use a number of providers around, but my ‘big’ servers are at OVH (which I really like since they are entirely hands-off on the software side of things, they only manage the hardware, and they’ve always told me when they are about to repair or fix or upgrade something and that there might be downtime but I’ve never had downtime with them), apparently you can VLAN your dedicated servers there. The ‘vRack’ costs a bit more but you only need one for any numbers of servers you want and you can tie in your servers elsewhere that are not hosted by OVH as well. It looks interesting, and could easily replace my SSH tunnel hacks, which is funny because it is apparently included for free with my largest server service… I should set it up sometime… ^.^;
outside of network, i would say Joyent ^^’
Doing your own networking on instant provisionning is hard because you need network control to offer that… Too tightly coupled…
About OpenStack : This is a nightmare and it does not work. It does replace a tons of internal shitty scripts you had, but automating shit producing just make it produce shit faster. On the long run, it rely as much as what was there before on the operators. So not really something that you can use.
Joyent is cool but pricing is about on par with AWS and it’s not really bare metal
IBM Bluemix IaaS (nee SoftLayer) offers bare metal hardware on a private VLAN. If you choose a fixed config server the provisioning won’t be “instant” but the servers are already racked and wired so your provision time depends largely on OS install. I’m not sure how close to your definition of “sane” the bandwidth pricing is though. (http://www.softlayer.com/bare-metal-servers)
Thank you looks interesting they still have high bandwidth pricing but it’s way better compared to AWS.
I should probably say, in the interest of full disclosure, that I work for IBM/SoftLayer.