[beagleboard] Re: Compiling the Linux kernel on Windows

One third option is to run Linux in a coLinux environment - I'm doing this, and it's actually working amazingly well, although compilation of a full image takes a *long* time... More info can be found at: http://www.colinux.org/

Best regards
  Søren

-----Oprindelig meddelelse-----

I second the recommendation. I'm doing the same thing, using a version of coLinux called "andLinux" ( http://www.andlinux.org/ ). Having my Linux windows intermixed with my regular Windows apps is very convenient.
The andLinux install is very easy, gives you full networking capability from inside Linux, and has a version of Ubuntu pre-installed, along with a lot of useful tools. And it is easy to add new tools, using standard apt-get mechanism. All of the data for the Linux side of things is stored in local "disk" files (one for each partition you have in Linux), and I've found it is very easy to make copies of these as either a backup of an entire Linux system state or to simply copy my entire dev setup to another PC in one shot.
The only issue I ran into was allowing external machines to access NFS services on the andLinux/coLinux side, and I worked that out eventually (and I'm happy to provide help if anyone else needs that).

Brett

Søren Steen Christensen wrote:

One third option is to run Linux in a coLinux environment -

> although compilation of a full image takes a *long* time...

Please define long. While it might be too late to change, I've been using a virtual PC, and it takes 24-48 hours depending on whether I'm doing anything on the host windows system. And that's for a fairly minimal image without X and related applications.

We did install linux on old pc, but it was actually much slower... ;^)

I did it a couple of weeks ago, and the BeagleBoard demo image took about
50-60 hours to compile (AFAIR) using OE on an Intel Core 2 6300@1.86GHz with
2 GB of memory, where coLinux was allowed to use 1GB of the memory.

coLinux only utilizes one core on a dual core/hyperthreading machine -
Unfortunately, although it leaves the other core(s) free for other work in
parallel, which as well is kind of nice :slight_smile:

Best regards
  Søren

-----Oprindelig meddelelse-----

Ouch... There are much better alternatives.

Simple Cross Compile avgs 15-30mins on P4 3.0Ghz
Native Build on Beagleboard running Debian Lenny: Averages 2 hours.
Native Build thru QEMU, on a Athlon X2 @ 2.6Ghz averaged 4-6 hours.
(8-10hrs when running two QEMU instances..)

Hi Robert,

15-30 min for going through all 6.300 (AFAIR) OE steps? My time as well
includes fetching all the stuff through a 2-4Mbps internet connection and
building the cross-tool-chain. But I totally agree - It isn't fast, though
possible :slight_smile:

Best regards
  Søren

-----Oprindelig meddelelse-----

sorry... just the kernel. :wink:

Robert Nelson wrote:

JustinLove <justin.love@cesgames.com> writes:

Robert Nelson wrote:

Hi Robert,

15-30 min for going through all 6.300 (AFAIR) OE steps? My time as well

sorry... just the kernel. :wink:

My time was based on 2400+ steps, if anybody is interested. The
kernel build alone is on the order of 30 minutes.

That's still a pretty long time. Over here a kernel build takes less
than 2 minutes.

Hi Måns,

On which kind of HW are you able to build the kernel (from scratch?) in just
2 minutes?
I think, that I have never had a kernel build (native or cross compiled)
below 10-15 min or so...

Are you running on any kind of
extreme-quad-core-super-deluxe-plus-extra-platform or doing anything special
in order to get this kind of performance? :slight_smile:

Best regards and thanks
  Søren

-----Oprindelig meddelelse-----

Søren Steen Christensen <sorenschristensen@stofanet.dk> writes:

My time was based on 2400+ steps, if anybody is interested. The
kernel build alone is on the order of 30 minutes.

That's still a pretty long time. Over here a kernel build takes less
than 2 minutes.

On which kind of HW are you able to build the kernel (from scratch?) in just
2 minutes?
I think, that I have never had a kernel build (native or cross compiled)
below 10-15 min or so...

Are you running on any kind of
extreme-quad-core-super-deluxe-plus-extra-platform or doing anything
special in order to get this kind of performance? :slight_smile:

I'm running a 2-year old Core2 Q6700 quad-core 2.66GHz. For extra
fun, I can use distcc with another dual-core machine as well.

Hi,

I think in this Linux will behave like any windows machine…
To increase speed first consider:

  • source file location
  • intermediate file location
  • outputfile location

on different disks (not partition,- I mean disks!) and make sure the system does not swap.
If you are using IDE/ATA also keep the disks on different IDE channels, IDE disks do not provide optimization of concurrent read/write operations on same channel as SAS or SCSI does.

Using a Raid will probably not improve performance, unless you have plenty of cache installed and write caching enabled.
Slow speed of applications with lots of disk I/O is usually introduced by positioning disk heads for input/output files and not really limited by transfer speed.

It is a very good Idea to keep intermediate files (working dir) on RAM-Disk, or at least SSD

It this doesn’t help any more, you may consider a processor upgrade,- but this would be last option for me…

regards,
AdT

Måns Rullgård schrieb: