5 ms·

All the 386-specific stuff seems like an argument in favor of "do things that don't scale."

breckinloggins 13y ago
Very much so. So many of my side projects get derailed due to thinking about scaling. I tend not to do web projects so "scaling" in my context is usually along the lines of "hmm, if I put this language feature in my compiler (that nobody knows about or uses and doesn't even work) I'm going to have to support it for years." Lame. The "do things that don't scale" article helped knock me out of this rut a bit. I suspect I'll knock myself out the rest of the way the more I remind myself that pre-mature worries like this are actually an advanced form of procrastination.
asveikau 13y ago
My memory of 1990s Linux development is hazy (I was a young kid at the time), but my recollection is ~5 years after this post they paid a huge cost in effort to get it to work on Alpha, and after that the required effort to port to other stuff went down. So maybe it's more like "simple initial implementation, massive rewrite later". Edit: Also worth noting that the 386 was the first "modern" Intel CPU in terms of instruction set and interfaces provided to kernel developers, so requiring 32-bit CPU with an MMU excludes a lot of machines in 1991 but not so many machines in the years that followed. What he's saying is he won't port to 286 which is a very obvious decision but maybe not so much if your frame of reference was Minix. (Later on I do seem to remember there used to be a port of Linux to MMU-less varieties of 68K in the late 90s, like some of the other replies are asking for. Not sure how well it ever worked.)
- cwzwarich 13y ago
  A variant of Linux for CPUs without MMUs lives on as http://www.uclinux.org/ http://www.uclinux.org/.
  - kristoffer 13y ago
    Actually since about 2.6 uCLinux has been merged into the mainline kernel. Several MMU less targets are supported.
aryastark 13y ago
The problem with generalizations is that once you go looking for them, you find them everywhere. Not everything is an affirmation of yet another Paul Graham paradigm. There is hardly anything more scalable than distributing a free software clone of a widely used OS over an open global network for cheap, incredibly popular, commodity hardware. If anything, MINIX did not scale. They charged a fee, and free Unix clones took off leaving MINIX behind.

Very much so. So many of my side projects get derailed due to thinking about scaling. I tend not to do web projects so "scaling" in my context is usually along the lines of "hmm, if I put this language feature in my compiler (that nobody knows about or uses and doesn't even work) I'm going to have to support it for years." Lame. The "do things that don't scale" article helped knock me out of this rut a bit. I suspect I'll knock myself out the rest of the way the more I remind myself that pre-mature worries like this are actually an advanced form of procrastination.

asveikau 13y ago

My memory of 1990s Linux development is hazy (I was a young kid at the time), but my recollection is ~5 years after this post they paid a huge cost in effort to get it to work on Alpha, and after that the required effort to port to other stuff went down. So maybe it's more like "simple initial implementation, massive rewrite later". Edit: Also worth noting that the 386 was the first "modern" Intel CPU in terms of instruction set and interfaces provided to kernel developers, so requiring 32-bit CPU with an MMU excludes a lot of machines in 1991 but not so many machines in the years that followed. What he's saying is he won't port to 286 which is a very obvious decision but maybe not so much if your frame of reference was Minix. (Later on I do seem to remember there used to be a port of Linux to MMU-less varieties of 68K in the late 90s, like some of the other replies are asking for. Not sure how well it ever worked.)

cwzwarich 13y ago
A variant of Linux for CPUs without MMUs lives on as http://www.uclinux.org/ http://www.uclinux.org/.
- kristoffer 13y ago
  Actually since about 2.6 uCLinux has been merged into the mainline kernel. Several MMU less targets are supported.

aryastark 13y ago

The problem with generalizations is that once you go looking for them, you find them everywhere. Not everything is an affirmation of yet another Paul Graham paradigm. There is hardly anything more scalable than distributing a free software clone of a widely used OS over an open global network for cheap, incredibly popular, commodity hardware. If anything, MINIX did not scale. They charged a fee, and free Unix clones took off leaving MINIX behind.