3 ms·

I think it's a little more nuanced than that. If I'm building something new, I generally want a new kernel/OS underneath. Especially if what I'm building can t

by awill 5y ago

I think it's a little more nuanced than that. If I'm building something new, I generally want a new kernel/OS underneath. Especially if what I'm building can take advantage of it (improvements to filesystems, networking stack etc..). After I've launched my service, I want stability over everything else. So here, I'd launch on the newest Ubuntu available, and stick with it for quite a long time.

moralestapia 5y ago
>I generally want a new kernel/OS underneath Huh? That's weird, Linux, OSX and Windows kernels have been battle-tested and hardened for like 30 years or so now. I can't imagine what kind of application would need a different kernel to start with. Aside from security concerns (which you didn't mention, plus one could argue an LTS is safer), could you give a specific example of: >what I'm building can take advantage of it (improvements to filesystems, networking stack etc..)
- orf 5y ago
  io_uring is one example, among many.
  - moralestapia 5y ago
    https://news.ycombinator.com/item?id=26745275 https://news.ycombinator.com/item?id=26745275 Is it really worth it to jump into a completely new feature to get marginal improvements on the most basic stuff (io, networking, etc...), all of this, in exchange of working with unstable APIs and unproven security vectors? My answer is no but I can picture people who would say yes. I guess it's a matter of personal preference (plus I work on stuff that has to be reliable, above everything else).
    - orf 5y ago
      It depends on what you’re building?
      - awill 5y ago
        yea, if you're building a distributed system that will need to scale to handle millions of TPS, then networking, IO, filesystem etc.. improvements can be significant.
- sbuttgereit 5y ago
  I read those comments as assuming new versions of those battle tested lineages... not newly created kernels or OSes. The last statement seems to make that intent clear.
  - moralestapia 5y ago
    I read those comments as "OP doesn't even know the difference between a kernel and a distro". The whole statement seems to nake that premise clear :^)

by awill 5y ago

moralestapia 5y ago

>I generally want a new kernel/OS underneath Huh? That's weird, Linux, OSX and Windows kernels have been battle-tested and hardened for like 30 years or so now. I can't imagine what kind of application would need a different kernel to start with. Aside from security concerns (which you didn't mention, plus one could argue an LTS is safer), could you give a specific example of: >what I'm building can take advantage of it (improvements to filesystems, networking stack etc..)

orf 5y ago
io_uring is one example, among many.
- moralestapia 5y ago
  https://news.ycombinator.com/item?id=26745275 https://news.ycombinator.com/item?id=26745275 Is it really worth it to jump into a completely new feature to get marginal improvements on the most basic stuff (io, networking, etc...), all of this, in exchange of working with unstable APIs and unproven security vectors? My answer is no but I can picture people who would say yes. I guess it's a matter of personal preference (plus I work on stuff that has to be reliable, above everything else).
  - orf 5y ago
    It depends on what you’re building?
    - awill 5y ago
      yea, if you're building a distributed system that will need to scale to handle millions of TPS, then networking, IO, filesystem etc.. improvements can be significant.
sbuttgereit 5y ago
I read those comments as assuming new versions of those battle tested lineages... not newly created kernels or OSes. The last statement seems to make that intent clear.
- moralestapia 5y ago
  I read those comments as "OP doesn't even know the difference between a kernel and a distro". The whole statement seems to nake that premise clear :^)