Building The Linux Kernel In 60 Seconds

[haplo602]

As said in the article, the actual review is coming up soon...

why isn't there a kernel compile time test in the 8150 test in the first place ? this would not start if you would stick to a standard test setup and not use tests not included there for anything.

I went through both BD tests (original one and compiler comparison) and no kernel compile time graph. don't try to refer me to openbenchmarking.org. the site has a horrible layour for me (completely unintuitive and unclear) and there are very few people that will actualy look there if they don't find information in phoronix.com articles.

[Michael]

why isn't there a kernel compile time test in the 8150 test in the first place ? this would not start if you would stick to a standard test setup and not use tests not included there for anything.

I went through both BD tests (original one and compiler comparison) and no kernel compile time graph. don't try to refer me to openbenchmarking.org. the site has a horrible layour for me (completely unintuitive and unclear) and there are very few people that will actualy look there if they don't find information in phoronix.com articles.

Only tests used in articles are those in the Phoronix Test Suite / OpenBenchmarking.org, which are all in a standardized configuration. The results are on OpenBenchmarking.org.

[del_diablo]

Only tests used in articles are those in the Phoronix Test Suite / OpenBenchmarking.org, which are all in a standardized configuration. The results are on OpenBenchmarking.org.

"I refuse to actually tell the truth, now your procrastination will make sure you won't find the result, so bugger off."

[Michael]

"I refuse to actually tell the truth, now your procrastination will make sure you won't find the result, so bugger off."

The compile time results of the FX-8150 and 7 other CPUs will also be in the 3960X review when published, if you don't feel like clicking around OpenBenchmarking.org to find them.

[FireBurn]

The compile time results of the FX-8150 and 7 other CPUs will also be in the 3960X review when published, if you don't feel like clicking around OpenBenchmarking.org to find them.

I sometimes wonder if you've tried using the site yourself to find results - I thought Gentoo had a steep learning curve

[Michael]

I sometimes wonder if you've tried using the site yourself to find results - I thought Gentoo had a steep learning curve

Search -> click -> Done. Feature requests and other constructive feedback is always welcome as well.

[ormaaj]

real    0m28.677s
user    1m25.066s
sys     0m0.000s

Full kernel build from make clean. Gotta love a ccache in RAM. Great for bisecting.

Here's my script to automate it. Only dependencies are ccache, pbzip2, bash4:

#!/usr/bin/env bash
# Copyright 2010, 2011 Dan Douglas
# License: WTFPL

declare -i csize=200                                      # max total cache size
cdir="/home/ormaaj/doc/mnt/ccache"                         # tmp directory
cachepath="/home/ormaaj/$(basename "$(pwd -P)").tar.bz2"   # cache directory
ccachedir="/usr/lib/ccache/bin"                           # ccache toolchain symlinks path

colorsetup() {
    set -- 'red' 'setaf 1' 'green' 'setaf 2' 'reset' 'sgr0'
    while (($#)); do
        colrs["$1"]="$(tput -S <<<"$2")"
        shift 2
    done

    echo() { printf -- '%s\n' "${colrs['green']}${@}${colrs['reset']}" ''; } >&2
}

die() {
    printf '%s\n' "${colrs['red']}${!#:-'Error occured.'}${colrs['reset']}" ''

    if [[ $1 == -n ]]; then
        echo 'Nonfatal, continuing...'
        return
    else
        local -i frame
        while caller $frame; do
            ((frame++))
        done
    fi

    exit 1
} >&2

cleanup() {
    echo 'Cleaning up tmpfs...'
    umount "$cdir" || die -n "Failed unmounting tmpfs at: ${cdir}"
}

doCcache() {
    if [[ ! -e $cdir ]]; then
        echo "No existing temporary cache directory ${cdir}, creating..."
        mkdir -p "$cdir"
    fi

    mount -t tmpfs -o "size=$((csize+50))m" tmpfs "$cdir" || die 'mount failed'

    trap -- 'cleanup' EXIT

    if [[ -e $cachepath ]]; then
        tar --use-compress-prog=pbzip2 -xvf "$cachepath" -C "$cdir" || die 'Failed unpacking cache files'
    else
        echo "No previous cache found at: ${cachepath}, creating new." >&2
    fi

    export PATH="${ccachedir:-'/usr/lib/ccache/bin'}:${PATH}" CCACHE_SIZE="$csize" CCACHE_DIR="$cdir"

    echo "$(ccache -s)"
    doCompile
    echo "$(ccache -s)"
    pushd "$cdir"
    tar --use-compress-prog=pbzip2 -cf "$cachepath" * || die -n 'Tar failed to pack up the ccache files!'
    popd
    cleanup
    trap - EXIT
}

doCompile() {
    make clean
    if make -j8; then
        make modules_install  || die
        make firmware_install || die
        cp "./arch/$(uname -m)/boot/bzImage" "/boot/$(basename "$(pwd -P)")" || die 'Failed while copying kernel image.'
    else
       die 'make failed'
    fi
}

main() {
    local -A colrs
    colorsetup
    pushd /usr/src/linux

    if [[ $1 == -n ]]; then
        doCompile
    else
        doCcache
    fi

#    schedtool -D -e emerge -v @module-rebuild
    popd
}

main "$@"

# vim: fenc=utf-8 ff=unix ts=4 sts=4 sw=4 ft=sh nowrap et :

[leeenux]

I think it's safe to say that Phoronix has sold out just like the other sites, as Michael has declined to be "open" and provide his kernel compiling parameters, but instead just invited us to stay tuned for his next article.

I compiled the kernel with

make defconfig
export CONCURRENCY_LEVEL=8
time make-kpkg --initrd --apend-to-version=-test kernel_image kernel_headers

On the following setup:

FX-8120
64GB Crucial SSD as boot drive, also where the kernel source was compiled
16GB of DDR3-1600

and compiled it in 4 minutes and 30 seconds. HOWEVER, the actual compilation took between 1.5 and 2 minutes with 90%+ CPU utilization across all cores the entire time, and the rest was those scripts that run before and after, where CPU and disk utilization are all well under 50% the entire time. This casts doubt on exactly how Michael is timing this benchmark, as I'm sure he can't be timing the entire make-kpkg command, which is what any logical person would consider a timed kernel compiling, and there's no way that the Intel CPU can somehow magically perform the non-CPU bound portion of the compile faster.

Of course, even if you compile the Linux kernel 10 times a day, the supposed, let's say: 10 extra minutes of waiting(assuming that we're comparing just the compilation part), is hardly going to inconvenience anybody, and I dare say that a $1000 16 core Opteron will do just as good and probably better than a $1000 Intel Xtreme edition CPU, not that anybody should be able to notice or care about it when the times are already so fast to begin with.

[Michael]

I think it's safe to say that Phoronix has sold out just like the other sites, as Michael has declined to be "open" and provide his kernel compiling parameters....

Huh? It's all open-source and there as part of the Phoronix Test Suite. Fetch the Phoronix Test Suite and run "phoronix-test-suite benchmark build-linux-kernel" and it will download, setup, and run the build test just like I do and then spit out your times.

[FunkyRider]

AMD has been a fscking loser for a long time, just like Mr Q