4 ms·

Tangential: TIL you can compile the Linux kernel in < 1 minute (on top-spec hardware). Seems it’s been a while since I’ve done that, because I remember it being

Tangential: TIL you can compile the Linux kernel in < 1 minute (on top-spec hardware). Seems it’s been a while since I’ve done that, because I remember it being more like an hour or more.

bornfreddy 1y ago
I remember I was blown away by some machine that compiled it in ~45 minutes. Pentium Pro baby! Those were the days.
- sgarland 1y ago
  My memory must be faulty, then, because I was mostly building it on an Athlon XP 2000+, which is definitely a few generations newer than a Pentium Pro. I’m probably thinking of various other packages, since at the time I was all-in on Gentoo. I distinctly remember trying to get distcc running to have the other computers (a Celeron 333 MHz and Pentium III 550 MHz) helping out for overnight builds. Can’t say that I miss that, because I spent more time configuring, troubleshooting, and building than using, but it did teach me a fair amount about Linux in general, and that’s definitely been worth it.
  - kelnos 1y ago
    Linux kernel compilation time depends heavily on what you're compiling though. You can have wildly different compilation times just by enabling or disabling some drivers or subsystems.
    - machomaster 1y ago
      Let's also not forget that kernel was not stagnant, but grew a lot over the years.
      - kelnos 1y ago
        Also true! Hard to say which was moving faster, though: CPU speed increases & RAM amount increase, or the additional code and complexity written into the kernel.
    - bpye 1y ago
      Yep - I have a 9950X desktop. Building the kernel as part of NixOS builds almost every possible option as a module - for ARM64 that takes something like 15 minutes.
- mwpmaybe 1y ago
  Painful memories of trying to build the kernel on my 486DX2/50, letting it run overnight and waking up to a compile-time error or non-booting kernel...
dahart 1y ago
Spinning hard drives were soooo slow! Maybe very roughly an order of magnitude from SSDs and an order of magnitude from multi-core?
- znpy 1y ago
  i think you nailed the centrail point. rotational disks usually top up at ~85 MB/sec, with seek time being up to 12 msec for consumer drives and ~6 msec for enterprise drives (15k rpm). ssd could saturate the sata bus and would top at 500-550 MB/sec, with essentially no seek time. latency wold be anything between 50 and 250 microseconds (depending on the operation). nvme disks instead can sometimes full utilise a pci-e lane and reach multiple gigabytes/second in sequential read (ie: pci-e gen5 nvme disks can peak at 7 GB/sec) with latencies as low as 10–30 microseconds for reads and 20–100 microseconds for writes. As compiling the kernel meant/means doing a lot of i/o on small files, you can see why disk access is a huge factor. A friend of mine did work on LLVM for his PhD... The first thing he did when he got funding for his phd was getting a laptop with as much memory as possible (i think it was 64gb on a dell workstation, at the time) and mount his work directory in tmpfs.
JonChesterfield 1y ago
I'd like to know why making debian packages containing the kernel now takes substantially longer than a clean build of the kernel. That seems deeply wrong and rather reduces the joy at finding the kernel builds so quickly.
- kelnos 1y ago
  Are you building the Debian kernel with the same .config as when you build the kernel directly? If not, that's probably why. A Debian kernel build will build basically everything, mostly as modules. A kernel tailored for your specific hardware and needs will be much faster to build. Also disabling building the _dbg package for the Debian build will significantly speed things up. Building that package takes a strangely long amount of time.
burnt-resistor 1y ago
In 2003, Egenera boxes could compile Linux in around 15 seconds. Even my dual 7402 with 96 threads and 512 GiB of RAM can't compile a maximal Linux config x86 build in RAM in under 3 minutes. What I find helps repeated builds is maintaining a memcache server instance for sccache.
mwpmaybe 1y ago
I suspect that's a very slimmed-down .config and probably only e.g. `make bzImage`, versus something like a Debian kernel .config with modules, headers, etc. A full Ubuntu kernel takes quite a bit longer on my 5950X, which is admittedly quite a bit slower than a 9950X but still no laggard. I'll time it and update this comment later... ETA: After dropping caches it took 18m40s to build Ubuntu mainline 6.16.3 with gcc-14 on my 5950X, with packages for everything including linux-doc, linux-rust-lib, linux-source, etc. I'd expect the same operation to take about 11m on a 9950X. It's also not clear whether OP used Clang or GCC to build the kernel. Clang would be faster.

Tangential: TIL you can compile the Linux kernel in < 1 minute (on top-spec hardware). Seems it’s been a while since I’ve done that, because I remember it being more like an hour or more.

bornfreddy 1y ago

I remember I was blown away by some machine that compiled it in ~45 minutes. Pentium Pro baby! Those were the days.

sgarland 1y ago
My memory must be faulty, then, because I was mostly building it on an Athlon XP 2000+, which is definitely a few generations newer than a Pentium Pro. I’m probably thinking of various other packages, since at the time I was all-in on Gentoo. I distinctly remember trying to get distcc running to have the other computers (a Celeron 333 MHz and Pentium III 550 MHz) helping out for overnight builds. Can’t say that I miss that, because I spent more time configuring, troubleshooting, and building than using, but it did teach me a fair amount about Linux in general, and that’s definitely been worth it.
- kelnos 1y ago
  Linux kernel compilation time depends heavily on what you're compiling though. You can have wildly different compilation times just by enabling or disabling some drivers or subsystems.
  - machomaster 1y ago
    Let's also not forget that kernel was not stagnant, but grew a lot over the years.
    - kelnos 1y ago
      Also true! Hard to say which was moving faster, though: CPU speed increases & RAM amount increase, or the additional code and complexity written into the kernel.
  - bpye 1y ago
    Yep - I have a 9950X desktop. Building the kernel as part of NixOS builds almost every possible option as a module - for ARM64 that takes something like 15 minutes.
mwpmaybe 1y ago
Painful memories of trying to build the kernel on my 486DX2/50, letting it run overnight and waking up to a compile-time error or non-booting kernel...

dahart 1y ago

Spinning hard drives were soooo slow! Maybe very roughly an order of magnitude from SSDs and an order of magnitude from multi-core?

znpy 1y ago
i think you nailed the centrail point. rotational disks usually top up at ~85 MB/sec, with seek time being up to 12 msec for consumer drives and ~6 msec for enterprise drives (15k rpm). ssd could saturate the sata bus and would top at 500-550 MB/sec, with essentially no seek time. latency wold be anything between 50 and 250 microseconds (depending on the operation). nvme disks instead can sometimes full utilise a pci-e lane and reach multiple gigabytes/second in sequential read (ie: pci-e gen5 nvme disks can peak at 7 GB/sec) with latencies as low as 10–30 microseconds for reads and 20–100 microseconds for writes. As compiling the kernel meant/means doing a lot of i/o on small files, you can see why disk access is a huge factor. A friend of mine did work on LLVM for his PhD... The first thing he did when he got funding for his phd was getting a laptop with as much memory as possible (i think it was 64gb on a dell workstation, at the time) and mount his work directory in tmpfs.

JonChesterfield 1y ago

I'd like to know why making debian packages containing the kernel now takes substantially longer than a clean build of the kernel. That seems deeply wrong and rather reduces the joy at finding the kernel builds so quickly.

kelnos 1y ago
Are you building the Debian kernel with the same .config as when you build the kernel directly? If not, that's probably why. A Debian kernel build will build basically everything, mostly as modules. A kernel tailored for your specific hardware and needs will be much faster to build. Also disabling building the _dbg package for the Debian build will significantly speed things up. Building that package takes a strangely long amount of time.

burnt-resistor 1y ago

In 2003, Egenera boxes could compile Linux in around 15 seconds. Even my dual 7402 with 96 threads and 512 GiB of RAM can't compile a maximal Linux config x86 build in RAM in under 3 minutes. What I find helps repeated builds is maintaining a memcache server instance for sccache.

mwpmaybe 1y ago

I suspect that's a very slimmed-down .config and probably only e.g. `make bzImage`, versus something like a Debian kernel .config with modules, headers, etc. A full Ubuntu kernel takes quite a bit longer on my 5950X, which is admittedly quite a bit slower than a 9950X but still no laggard. I'll time it and update this comment later... ETA: After dropping caches it took 18m40s to build Ubuntu mainline 6.16.3 with gcc-14 on my 5950X, with packages for everything including linux-doc, linux-rust-lib, linux-source, etc. I'd expect the same operation to take about 11m on a 9950X. It's also not clear whether OP used Clang or GCC to build the kernel. Clang would be faster.