6 ms·

Right. All I need from a node is the CPU / Memory and a docker daemon. The less I know about an OS or any other services / patches, etc the better. I.e. I want

Right. All I need from a node is the CPU / Memory and a docker daemon. The less I know about an OS or any other services / patches, etc the better. I.e. I want my nodes to be cattles.

beagle3 7y ago
The same mindset would dictate that you only ever buy no-name hardware which is about 30% cheaper than dell/hp for the same performance characteristics... why would anyone do that? And that's why people DO pay for red-hat. Peace of mind, whether supported by evidence or not.
- streetcat1 7y ago
  So people pay for redhat because they did not move to container/ microservices architecture. Any main stream technology (Mainframe, Dec/Vax, PC, VM) will always have legacy customers. However, one the state of the art architecture is micro services, the optimal runtime is kuberentes. Once everyone is on kubernetes hardware will not matter. Moreover, I would love a cloud based on second hand hardware (which should be much cheaper than 30%).
  - snuxoll 7y ago
    > Moreover, I would love a cloud based on second hand hardware (which should be much cheaper than 30%) I’ve wanted to do this for some time, but there’s a lot of cost involved in the space (square footage itself, electricity, cooling) that you can’t cut just by running older stuff. Bandwidth and storage cost the same regardless as well (don’t buy used disks for anything remotely important).
    - mbreese 7y ago
      An even bigger cost is power (if we’re talking about a lot of compute capacity- eg a private cloud). A cluster that’s running older servers might be 30% cheaper to purchase, but it will likely be more power inefficient, which would make it significantly more expensive to actually operate. Depending on your workload, it is often the case that purchasing new (more efficient) hardware can actually be the cheaper option than running on older hardware from an OpEx standpoint.
      - snuxoll 7y ago
        This part can vary a lot these days at least, the Ivy/Sandy Bridge xeon chips are still remarkably capable and I can buy a used 11th gen Dell server for a couple hundred dollars. Before the Westmere/Nehalam chips it was a different story, but anything newer is going to sip power and perform well as long as you know you’re not getting “all the perf out of it”. It’s entirely possible to make a business out of this, but you rely on a customer base that’s inherently cheap and fussy by nature. Not all of them are, but most looking “for a deal” are cheapskates who will milk you dry in time/labor if you let them and you don’t have the margins to deal with that bullshit.
  - ineedasername 7y ago
    There's still a very large enterprise market for monolithic software that isn't going anywhere anytime soon. These are not legacy customers.
  - beagle3 7y ago
    What is the hypervisor OS? RedHat certifies hardware so you know your disk controllers and hot-swappable memory and CPU are supported. You CAN use Alpine as the hypervisor, but IBM/RedHat/Oracle customers typically don't.
z3t4 7y ago
What you don't know can't hurt you :P Abstraction layers are good when it feels that way though.

Right. All I need from a node is the CPU / Memory and a docker daemon. The less I know about an OS or any other services / patches, etc the better. I.e. I want my nodes to be cattles.

beagle3 7y ago

The same mindset would dictate that you only ever buy no-name hardware which is about 30% cheaper than dell/hp for the same performance characteristics... why would anyone do that? And that's why people DO pay for red-hat. Peace of mind, whether supported by evidence or not.

streetcat1 7y ago
So people pay for redhat because they did not move to container/ microservices architecture. Any main stream technology (Mainframe, Dec/Vax, PC, VM) will always have legacy customers. However, one the state of the art architecture is micro services, the optimal runtime is kuberentes. Once everyone is on kubernetes hardware will not matter. Moreover, I would love a cloud based on second hand hardware (which should be much cheaper than 30%).
- snuxoll 7y ago
  > Moreover, I would love a cloud based on second hand hardware (which should be much cheaper than 30%) I’ve wanted to do this for some time, but there’s a lot of cost involved in the space (square footage itself, electricity, cooling) that you can’t cut just by running older stuff. Bandwidth and storage cost the same regardless as well (don’t buy used disks for anything remotely important).
  - mbreese 7y ago
    An even bigger cost is power (if we’re talking about a lot of compute capacity- eg a private cloud). A cluster that’s running older servers might be 30% cheaper to purchase, but it will likely be more power inefficient, which would make it significantly more expensive to actually operate. Depending on your workload, it is often the case that purchasing new (more efficient) hardware can actually be the cheaper option than running on older hardware from an OpEx standpoint.
    - snuxoll 7y ago
      This part can vary a lot these days at least, the Ivy/Sandy Bridge xeon chips are still remarkably capable and I can buy a used 11th gen Dell server for a couple hundred dollars. Before the Westmere/Nehalam chips it was a different story, but anything newer is going to sip power and perform well as long as you know you’re not getting “all the perf out of it”. It’s entirely possible to make a business out of this, but you rely on a customer base that’s inherently cheap and fussy by nature. Not all of them are, but most looking “for a deal” are cheapskates who will milk you dry in time/labor if you let them and you don’t have the margins to deal with that bullshit.
- ineedasername 7y ago
  There's still a very large enterprise market for monolithic software that isn't going anywhere anytime soon. These are not legacy customers.
- beagle3 7y ago
  What is the hypervisor OS? RedHat certifies hardware so you know your disk controllers and hot-swappable memory and CPU are supported. You CAN use Alpine as the hypervisor, but IBM/RedHat/Oracle customers typically don't.

z3t4 7y ago

What you don't know can't hurt you :P Abstraction layers are good when it feels that way though.