Agreed. I would just move on to the next article, except that there is a need here for a company I invest in. The company currently manages thousands of hosts and the cloud is eating into their profits. The hardware out there is way to expensive (or a bad job was done searching) to move away from the cloud. If they can provide good hardware that would allow the thousands of host company to save money if they did it on their own, then we would be interested in looking into it.
Off the top of my head of what would be required besides just cheap hardware.
1) Cloud like software to manage it or k8s native support. I don't know exactly how that would be done, but the administrative costs of using this hardware can't be so high as to make the cloud a more viable option.
2) Some options for network access. Cloud does not just provide VM's but the underlying reliable network with multiple redundant pipes. Comcast Internet access might be good enough, but some customers might require large redundant pipes. While they might NOT need to solve this themselves, they should make sure that the market does provide solutions that when taken into account, allows for cheaper than cloud solutions.
3) Physical location... same as item 2 but for physical location of the hardware.
I remember back in the day managing colos. It really sucked. It was not just hardware costs that sucked. They should just consider that.
> I would just move on to the next article, except that there is a need here for a company I invest in.
This is exactly how any "WeWork" type company sucks in investors. You'd just ignore the stupid out of hand except that it would be so nice to believe that if this company is on the level and IF their product does what they say it will and IF they deliver it in a reasonable time frame then it will solve the problem and somehow make money for investors.
Forget the names and reputations of the people involved... if three random people came up to you at a conference and said "We've formed a company to solve problem X!", wouldn't you wait until they showed an actual product to even think about potentially making decisions based on what they might do?
I think providing reliable compute and storage is not so difficult and for that there are many options that are cheaper than AWS or Azure (e.g. Hetzner cloud, Scaleway, Vultr or DO). The hard part, in my opinion, is providing managed services like relational databases or key-value stores. For that there are far fewer vendors available, so if Oxide manages to build something that makes this easier I think they’d have a pretty solid business case. In general I think we see a trend towards simpler IT and system architectures, which I really like as I think the complexity has become way too high. I e.g. know a company that invested significant resources (people and hardware) in setting up an OpenStack cluster but even after three years never managed to use it in production. From what I’ve heard k8s has a similar complexity problem, so I really think there’s room for a simpler approach to scalable computing.
Honestly I feel that far too many companies are afraid of building their own servers. In addition to getting exactly what you need for each workload, it can save the company a boatload of money over time. System integrators like Dell charge companies millions of dollars to assemble servers that are no better than anything I could build myself.
Another overlooked benefit of building your own hardware is you don’t need to add complex overhead like virtualization or containers. You can “right size” each server with the perfect amount of CPU, IO and RAM for the processes you will run on it. Paired with a kernel compiled specifically for that server, your businesses applications will be vastly more performant than ones running on a cloud platform.
Honestly, too many businesses hop on the latest fads. The smart ones see the value in going back to the basics.
If SuperMicro plus real estate, power, provisioning, and operating costs isn't competitive with cloud costs, perhaps it might be because the large cloud vendors have entire teams focusing entirely on bringing down the cost of building and operating these things at scale.
Cloud vendors don't have 90% profit margins; the cost of cloud is what it is because operating computers at scale is hard.
Off the top of my head of what would be required besides just cheap hardware. 1) Cloud like software to manage it or k8s native support. I don't know exactly how that would be done, but the administrative costs of using this hardware can't be so high as to make the cloud a more viable option. 2) Some options for network access. Cloud does not just provide VM's but the underlying reliable network with multiple redundant pipes. Comcast Internet access might be good enough, but some customers might require large redundant pipes. While they might NOT need to solve this themselves, they should make sure that the market does provide solutions that when taken into account, allows for cheaper than cloud solutions. 3) Physical location... same as item 2 but for physical location of the hardware.
I remember back in the day managing colos. It really sucked. It was not just hardware costs that sucked. They should just consider that.