4 ms·

The wording is certainly interesting. This article is written as if cloud was a holy grail, with companies still being stuck in the stone-age era of.... owning

The wording is certainly interesting. This article is written as if cloud was a holy grail, with companies still being stuck in the stone-age era of.... owning servers. They also pretend that the cloud is the future for everything, and that on-premise servers are legacy technology. The following paragraph is the most illuminating: "The decision to build or rent may depend on a user’s data intensity. Consider a firm in America with a medium-sized data centre which thinks its computing load will rise roughly four-fold over a decade. In that case, building its own breaks even in seven years or so, and ends up being 5% cheaper overall than leasing, according to data from Schneider Electric. If the load remains stable, renting is the less expensive option, by a similar amount (assuming a flat cost of capital)." This assumes that cloud costs are stagnant and reliable. This is a very naive assumption - it can be seen that if the cloud providers think they can get away with cost increases, they will. On-premises works out cheaper in the majority of cases, even including the cost of a sysadmin team.

VirusNewbie 3y ago
I’ve never worked at a small company when on-prem is cheaper, because having to provision for peak load is not cost effective.
- jeremyjh 3y ago
  A lot of small companies have very stable workloads for years. Unless you meant startups?
- johnklos 3y ago
  That's a sales line. Or do you work exclusively for small companies that are selling high demand online services that're trying really hard to go viral? The only other time I can imagine is if a company isn't going to exist in six months, such as movie productions, or companies that have so little going on that they don't need more than a $10 a month plan. Otherwise, on premises or colocation are invariably cheaper when there's any kind of real usage.
  - VirusNewbie 3y ago
    I’ve worked for a small telecom where load was variable, new contracts sometimes meant double or triple the capacity, other times we had smaller contracts without bursty load. The other company was software for automotive companies and marketing, only US based so traffic was non existent from midnight to 5:00 am, and some companies would need to ramp up for events like NFL games when they’d advertise.
- safetytrick 3y ago
  It just depends on the business, b2b can be very consistent. Some workloads just have predictable behavior. I worked at a medium sized place where we had plans in place to handle the workload for our entire market on-prem (if we were ever to capture that market). We realized that horizontal scalability wasn't important to us because we had enough of the market to know that the other half of the market could be supported on our existing solution.
  - VirusNewbie 3y ago
    If your market is regional and not worldwide, you almost assuredly have diurnal load spikes and periods of time where you have almost no activity.
    - safetytrick 3y ago
      Some services just don't stop at night. Healthcare has trends but it never stops.
- downrightmike 3y ago
  It becomes cheaper because they will run long after support has ended and cobble together a working server from the pile of broken ones, all the while rejecting getting a new server that could handle and replace dozens.
- jjav 3y ago
  > having to provision for peak load is not cost effective. That is true, if peak load is much much higher than mean load. For most companies, that's not the case. On the cloud you are overpaying for cpu, network and storage by a very wide margin. That makes sense if your peak load is, say, 100x your mean load (and you have sufficienly smart autoscaling to optimize for this). But for most companies there is not such huge multiplier from mean usage to peak usage. So those are better off with owning hardware.
fatherzine 3y ago
datacenter is to information what power plant is to electricity Economically, scale makes the centralized solution more efficient. That doesn't preclude independent entities running their own stoves, or running their own internal combustion engines, or their own industrial furnaces, but confines those uses to niche areas where direct power generation is more convenient/efficient due to additional constraints. Politically, centralization means power. Power seeking entities will eventually outright ban independent power generation, for whatever excuse du jour. See NY ban of gas stoves, or EU plans to ban ICE cars. For information, 'to prevent illegal activity' excuse is usually used.
- jjav 3y ago
  > datacenter is to information what power plant is to electricity Very much not. Running a generator at home is much more expensive than paying your utility for electricity. But running a server at home is far cheaper than paying AWS if you actualy use the server.
Adiqq 3y ago
I don't have much contact with cloud, but it just seems expensive and unstable in practice from business perspective. Cloud offers standardization and support, so in some cases it might be preferable, because it meets requirements without spending additional time on setup or operations, but eventually cost might become problem, especially if cloud providers will decide to increase prices and you can't easily migrate it to on-premise. Probably it's best to stay hybrid and be always ready to at least change cloud providers, so you need to avoid vendor lock-in. That sounds like cloud native, so why bother with cloud, if you can have some servers, use them with Rancher and create Kubernetes clusters? There will be some overhead, but your company controls servers, can easily migrate also to cloud, if needed, provides most functionality that most teams may require and depends only on electricity, internet providers and workers required to operate some data center, so it can always choose cheapest option to operate.
renewiltord 3y ago
Big problem with on-prem is lead time on expansion. Can easily slow down expansion. Hard to project out. But if you have high utilization stable load, it can pay off to have on-prem. It's still a big risk, though. Cloud has a stable platform to build off. You can trust the 'interface' so to speak. Meanwhile you suddenly have to worry about driver incompatibilities, disk wear, stuff like that. We have cloud infra and on-prem infra. Big problem with cloud is performance for price. Even for small company with stable load, think about it. You have to have a guy aware of and thinking of the fact that the Mellanox SFP+ card you have has a driver regression on Linux 6.2. Maintenance nightmare. Better to have cloud. Clear interface to build against, and easy to replicate when problems occur. Sysadmins do stuff. You can't go sysadmin-free and expect efficient spending. Someone else just ends up sysadmin and frequently someone too high paid so you're losing money on the time they play that role.
jjav 3y ago
Cloud makes economic sense in two scenarios. For startups that don't have a a clue what they will need because still searching for market fit so it doesn't make sense to provision hardware not knowing what is needed. If you suddenly find market fit and usage grows 100x+ in a few months, you totally want to be on cloud. And in larger coporations, those that have extremely spiky usage. For example a company that has 90% of their usage at the end of each month, it doesn't make sense to own hardware to meet the peak load if it will be idle 28 days of the month. Most corporations are not in those two categories though! If you are not, you are vastly overpaying for compute, storage and network compared to owning your own hardware. It is sad that such companies are rushing to overpay for cloud just because it's hip. Objectively analyze your needs and usage and make a data-driven decision. Don't rush into cloud just because everyone else did.

VirusNewbie 3y ago

I’ve never worked at a small company when on-prem is cheaper, because having to provision for peak load is not cost effective.

jeremyjh 3y ago
A lot of small companies have very stable workloads for years. Unless you meant startups?
johnklos 3y ago
That's a sales line. Or do you work exclusively for small companies that are selling high demand online services that're trying really hard to go viral? The only other time I can imagine is if a company isn't going to exist in six months, such as movie productions, or companies that have so little going on that they don't need more than a $10 a month plan. Otherwise, on premises or colocation are invariably cheaper when there's any kind of real usage.
- VirusNewbie 3y ago
  I’ve worked for a small telecom where load was variable, new contracts sometimes meant double or triple the capacity, other times we had smaller contracts without bursty load. The other company was software for automotive companies and marketing, only US based so traffic was non existent from midnight to 5:00 am, and some companies would need to ramp up for events like NFL games when they’d advertise.
safetytrick 3y ago
It just depends on the business, b2b can be very consistent. Some workloads just have predictable behavior. I worked at a medium sized place where we had plans in place to handle the workload for our entire market on-prem (if we were ever to capture that market). We realized that horizontal scalability wasn't important to us because we had enough of the market to know that the other half of the market could be supported on our existing solution.
- VirusNewbie 3y ago
  If your market is regional and not worldwide, you almost assuredly have diurnal load spikes and periods of time where you have almost no activity.
  - safetytrick 3y ago
    Some services just don't stop at night. Healthcare has trends but it never stops.
downrightmike 3y ago
It becomes cheaper because they will run long after support has ended and cobble together a working server from the pile of broken ones, all the while rejecting getting a new server that could handle and replace dozens.
jjav 3y ago
> having to provision for peak load is not cost effective. That is true, if peak load is much much higher than mean load. For most companies, that's not the case. On the cloud you are overpaying for cpu, network and storage by a very wide margin. That makes sense if your peak load is, say, 100x your mean load (and you have sufficienly smart autoscaling to optimize for this). But for most companies there is not such huge multiplier from mean usage to peak usage. So those are better off with owning hardware.

fatherzine 3y ago

datacenter is to information what power plant is to electricity Economically, scale makes the centralized solution more efficient. That doesn't preclude independent entities running their own stoves, or running their own internal combustion engines, or their own industrial furnaces, but confines those uses to niche areas where direct power generation is more convenient/efficient due to additional constraints. Politically, centralization means power. Power seeking entities will eventually outright ban independent power generation, for whatever excuse du jour. See NY ban of gas stoves, or EU plans to ban ICE cars. For information, 'to prevent illegal activity' excuse is usually used.

jjav 3y ago
> datacenter is to information what power plant is to electricity Very much not. Running a generator at home is much more expensive than paying your utility for electricity. But running a server at home is far cheaper than paying AWS if you actualy use the server.

Adiqq 3y ago

I don't have much contact with cloud, but it just seems expensive and unstable in practice from business perspective. Cloud offers standardization and support, so in some cases it might be preferable, because it meets requirements without spending additional time on setup or operations, but eventually cost might become problem, especially if cloud providers will decide to increase prices and you can't easily migrate it to on-premise. Probably it's best to stay hybrid and be always ready to at least change cloud providers, so you need to avoid vendor lock-in. That sounds like cloud native, so why bother with cloud, if you can have some servers, use them with Rancher and create Kubernetes clusters? There will be some overhead, but your company controls servers, can easily migrate also to cloud, if needed, provides most functionality that most teams may require and depends only on electricity, internet providers and workers required to operate some data center, so it can always choose cheapest option to operate.

renewiltord 3y ago

Big problem with on-prem is lead time on expansion. Can easily slow down expansion. Hard to project out. But if you have high utilization stable load, it can pay off to have on-prem. It's still a big risk, though. Cloud has a stable platform to build off. You can trust the 'interface' so to speak. Meanwhile you suddenly have to worry about driver incompatibilities, disk wear, stuff like that. We have cloud infra and on-prem infra. Big problem with cloud is performance for price. Even for small company with stable load, think about it. You have to have a guy aware of and thinking of the fact that the Mellanox SFP+ card you have has a driver regression on Linux 6.2. Maintenance nightmare. Better to have cloud. Clear interface to build against, and easy to replicate when problems occur. Sysadmins do stuff. You can't go sysadmin-free and expect efficient spending. Someone else just ends up sysadmin and frequently someone too high paid so you're losing money on the time they play that role.

jjav 3y ago

Cloud makes economic sense in two scenarios. For startups that don't have a a clue what they will need because still searching for market fit so it doesn't make sense to provision hardware not knowing what is needed. If you suddenly find market fit and usage grows 100x+ in a few months, you totally want to be on cloud. And in larger coporations, those that have extremely spiky usage. For example a company that has 90% of their usage at the end of each month, it doesn't make sense to own hardware to meet the peak load if it will be idle 28 days of the month. Most corporations are not in those two categories though! If you are not, you are vastly overpaying for compute, storage and network compared to owning your own hardware. It is sad that such companies are rushing to overpay for cloud just because it's hip. Objectively analyze your needs and usage and make a data-driven decision. Don't rush into cloud just because everyone else did.