5 ms·

My business just started self-hosting, but we severely underestimated the cost of air conditioning (installation), SFP adapters, enterprise Ethernet and electri

My business just started self-hosting, but we severely underestimated the cost of air conditioning (installation), SFP adapters, enterprise Ethernet and electrical infrastructure and ended up spending as much on peripheral stuff as servers. So far the whole operation has cost around $30k (small business). We stand to save $1-1.5k per month at the moment, less the cost of symmetrical fiber about $500/mo. Next year it will pay off but this year we definitely won’t break even on the servers. How has this has changed our culture? Well, what I have enjoyed about the whole process is now we have a bunch of spare capacity, so we can spin up demo servers for SMB clients for even a few months and it “costs us nothing”. Then we can arrange payment later when they fully appreciate the service. We don’t take any risks anymore with bills that need to be settled to cloud providers. This has changed how “generous” we are in the outset with our clients. It feels really healthy to be able to do this.

javchz 5y ago
I'm glad you did. One of the things that scare me the most about the current state of the Internet, it's how centralized it's everything in the "cloud" with a lot of the Internet infrastructure depending on like 5 providers. Self hosting it's dirty, time consuming, expensive, risky, but it's predictable. The problem with XaaS it's how little control we have. Not only the rules of pricing can change any day, but if you're already dependable on the service, they could easily shut down that service if it's not profitable enough, and with some luck you will only be able to migrate in a few months.
- sdoering 5y ago
  In the end I think it will boil down to a question of anti fragility or resiliency VS. efficiency. And there is actually imho no black and white answer. Take a current, very small freelance client of mine. I built them a homepage based on Kirby CMS. Hosted on uberspace.de. On the same shared server there is Matomo being hosted for basic stats. Another client I was involved in went another way. They had their site built by a team of designers with webflow and are using Google Analytics. In both cases the decision makes sense given the conditions and constraints of the respective projects. Personally I would have not recommend to use webflow, but that was not my call to make. For GA I provided a pro/con evaluation and the client decided to go with GA (not my primary recommendation). I still think for his use case at the time a sensible decision.
me_me_mu_mu 5y ago
Dumb question. I’m building little social apps, and I’m hosting stuff in GCP or AWS. I have a few old computers that are still kinda beefy (quad cores and a 6-core, 16+ Gb ram, some old but still working 3TB hdd). Would it be stupid to run some of my code and data on these? Not everything I have needs to be geographically close to users or something like that, such as a mass notifier service that runs every so often. For example I’ve got a few auxiliary services that I’m sure I could run just fine off these machines, but I chose the cloud because it’s what I’m used to from work.. I have a fiber connection at home and I’m anyway powering my laptops and other electronics 24/7.. Obviously not a bigger business like your use case but I’m just curious. Some services cost quite a bit and I feel like I have better hardware for some things (non database) than what I pay for.
- comprev 5y ago
  I would recommend plugging the hardware into power measuring units before the wall sockets to give you an estimated running cost. Depending on where you are in the world electricity can be crazy expensive. The cost of purchasing a lower power SBC (x86, ARM) might be very quickly offset. A RaspberryPi can handle a surprising amount, especially the 4/8GB models. The trick is using low resource software.
- q1w2 5y ago
  You aren't really paying for computing power and storage - you are paying for power/network/facility/server redundancy and availability. We host out own servers, and the bulk of the time is ensuring that the power doesn't go out, that there is a backup internet connection, that the HDs are on RAID, that the server has a backup server available (and the router, and the switch, and the NICs), and that the management network is available, and that the location has proper security monitoring, and that the offsite backup happens nightly, and that we have an employee that knows how to do all of the above... All of that is time consuming and expensive. ...BUT, some people/some companies actually enjoy that sort of thing, in which case, it can work out financially - though often you're sacrificing one of the 9's in your 99.99% uptime/reliability.
  - horsawlarway 5y ago
Sammi 5y ago
What I don't get is if the problem was the risk of unexpected bills from a cloud provider, then why not just use simple virtual servers that you can rent for a fixed price from almost any hosting provider? Gives you all that extra benefit you speak of of the extra compute headroom that you can use for spinning up any demo server application you want, and saves you all the costs, risks, and headache of managing your own hardware. Why the debate of cloud vs on prem? The logical sweet spot in the middle seems to be vps to me.
- horsawlarway 5y ago
  I mean, I see VPS as the worst of both worlds. You're still managing a server, but now you're paying through the nose to do it. The costs just simply don't work out nicely to anyone who can afford to buy at even small scales. A 4 cpu, 8gb ram machine will pretty commonly run you $50 a month, with less than 200gb storage. You're paying 600/year for a machine that I can build (with better specs) for ~400. Not to mention that the machine you buy yourself will have a lifespan of 3 to 8 years. So sure - once you factor in bandwidth, storage location, power - you're probably not going to save money in the first year. But by year 2 you are, and by year 4 you've saved a considerable sum.
  - aetherspawn 5y ago
    My thoughts are same as above, and also -- if the app that you're hosting only runs on Windows -- then Windows VPS are really ridiculously expensive. Some providers charge around $30/mo for a key. You save a lot of money being able to buy/manage your own Windows keys. ATM we're running Windows under KVM and the performance is great anyway. But if you have to run a lot of Windows boxes, it seems like the cost effective way is to buy a Windows Server DC license ... then all the guest VM licenses are free, and with the right server hardware you can probably cram like 20 VMs on a single box.

javchz 5y ago

I'm glad you did. One of the things that scare me the most about the current state of the Internet, it's how centralized it's everything in the "cloud" with a lot of the Internet infrastructure depending on like 5 providers. Self hosting it's dirty, time consuming, expensive, risky, but it's predictable. The problem with XaaS it's how little control we have. Not only the rules of pricing can change any day, but if you're already dependable on the service, they could easily shut down that service if it's not profitable enough, and with some luck you will only be able to migrate in a few months.

sdoering 5y ago
In the end I think it will boil down to a question of anti fragility or resiliency VS. efficiency. And there is actually imho no black and white answer. Take a current, very small freelance client of mine. I built them a homepage based on Kirby CMS. Hosted on uberspace.de. On the same shared server there is Matomo being hosted for basic stats. Another client I was involved in went another way. They had their site built by a team of designers with webflow and are using Google Analytics. In both cases the decision makes sense given the conditions and constraints of the respective projects. Personally I would have not recommend to use webflow, but that was not my call to make. For GA I provided a pro/con evaluation and the client decided to go with GA (not my primary recommendation). I still think for his use case at the time a sensible decision.

me_me_mu_mu 5y ago

Dumb question. I’m building little social apps, and I’m hosting stuff in GCP or AWS. I have a few old computers that are still kinda beefy (quad cores and a 6-core, 16+ Gb ram, some old but still working 3TB hdd). Would it be stupid to run some of my code and data on these? Not everything I have needs to be geographically close to users or something like that, such as a mass notifier service that runs every so often. For example I’ve got a few auxiliary services that I’m sure I could run just fine off these machines, but I chose the cloud because it’s what I’m used to from work.. I have a fiber connection at home and I’m anyway powering my laptops and other electronics 24/7.. Obviously not a bigger business like your use case but I’m just curious. Some services cost quite a bit and I feel like I have better hardware for some things (non database) than what I pay for.

comprev 5y ago
I would recommend plugging the hardware into power measuring units before the wall sockets to give you an estimated running cost. Depending on where you are in the world electricity can be crazy expensive. The cost of purchasing a lower power SBC (x86, ARM) might be very quickly offset. A RaspberryPi can handle a surprising amount, especially the 4/8GB models. The trick is using low resource software.
q1w2 5y ago
You aren't really paying for computing power and storage - you are paying for power/network/facility/server redundancy and availability. We host out own servers, and the bulk of the time is ensuring that the power doesn't go out, that there is a backup internet connection, that the HDs are on RAID, that the server has a backup server available (and the router, and the switch, and the NICs), and that the management network is available, and that the location has proper security monitoring, and that the offsite backup happens nightly, and that we have an employee that knows how to do all of the above... All of that is time consuming and expensive. ...BUT, some people/some companies actually enjoy that sort of thing, in which case, it can work out financially - though often you're sacrificing one of the 9's in your 99.99% uptime/reliability.
- horsawlarway 5y ago

Sammi 5y ago

What I don't get is if the problem was the risk of unexpected bills from a cloud provider, then why not just use simple virtual servers that you can rent for a fixed price from almost any hosting provider? Gives you all that extra benefit you speak of of the extra compute headroom that you can use for spinning up any demo server application you want, and saves you all the costs, risks, and headache of managing your own hardware. Why the debate of cloud vs on prem? The logical sweet spot in the middle seems to be vps to me.

horsawlarway 5y ago
I mean, I see VPS as the worst of both worlds. You're still managing a server, but now you're paying through the nose to do it. The costs just simply don't work out nicely to anyone who can afford to buy at even small scales. A 4 cpu, 8gb ram machine will pretty commonly run you $50 a month, with less than 200gb storage. You're paying 600/year for a machine that I can build (with better specs) for ~400. Not to mention that the machine you buy yourself will have a lifespan of 3 to 8 years. So sure - once you factor in bandwidth, storage location, power - you're probably not going to save money in the first year. But by year 2 you are, and by year 4 you've saved a considerable sum.
- aetherspawn 5y ago
  My thoughts are same as above, and also -- if the app that you're hosting only runs on Windows -- then Windows VPS are really ridiculously expensive. Some providers charge around $30/mo for a key. You save a lot of money being able to buy/manage your own Windows keys. ATM we're running Windows under KVM and the performance is great anyway. But if you have to run a lot of Windows boxes, it seems like the cost effective way is to buy a Windows Server DC license ... then all the guest VM licenses are free, and with the right server hardware you can probably cram like 20 VMs on a single box.