Slow PXE boot

[Domagoj Smolčić]

Mark,

PXE clients are directly connected to FreeBSD based PXE server, with same ETH cable.
There is no any intermediate hardware like switch, etc ...
Regarding '... running at 100mb 1/2 duplex', nope ..., in bios setting is all set to "max" for NICs

Finally, result is always 100% repeatable for both tested clients.
One always being super fast and the other, unbelievably slow ...


On Mon, 24 Feb 2020 08:46:20 -0500
Mark Saad <nonesuch at longcount.org> wrote:

> Domagoj 
>    I regularly boot 11.3-Stable and 12.1-Stable amd64 on servers with
> 64gb, 128gb , and occasional more . I do t have that issue . Can you
> take a peak at the switch your boxes are attached to ; It sounds like
> a device is running at 100mb 1/2 duplex . I remember the realtec nics
> doing this when set to auto . 
> 
> ---
> Mark Saad | nonesuch at longcount.org
> 
> > On Feb 23, 2020, at 9:23 PM, Domagoj Smolčić
> > <rank1seeker at gmail.com> wrote:
> > 
> > Yo Crew!
> > 
> > During PXE boot, from server (11.3-RELEASE-p5 i386) with 1 GB RAM,
> > client with: 4 GB RAM booted kernel in  ~2 min
> >    8 GB RAM booted kernel in ~20 min (at least)
> > So all below is tested with 2 different clients, each with NIC of
> > different manufacturer (Realtec & Intel)
> > 
> > I've read on forums that if client has much, much more RAM than
> > server, this happens. Suggestion was to to set loader(8) tunable
> > hw.memtest.tests to 0, so client would skip RAM test.
> > 
> > PXE boot is served to clients from /PXE, so ...
> > Setting hw.memtest.tests="0" in /PXE/boot/loader.conf, yielded no
> > results!
> > 
> > By the way, for some reason, hw.memtest.tests tunable DOESN'T exist
> > in man pages! I had to look in source code to confirm it still
> > exists.
> > 
> > 
> > Next, I've tried:
> > # echo 'nfs.read_size="16384"' > /PXE/boot/loader.conf
> >    Changing value to ANY value other than default 1024, completely
> > halts client's kernel loading at 'Loading kernel...'!
> > 
> > 
> > Lastly, I've tried tactic used with UFS's stage 2 /boot/boot, to
> > skip stage 3 loader completely and to directly load kernel instead.
> > So in dhcpd.conf, I've replaced: filename "boot/pxeboot";
> > with:
> >    filename "boot/kernel/kernel";
> > 
> > Result:
> >    Attempt to pull GENERIC kernel directly, resulted in: "NBP is
> > too big to fit in free base memory" Maybe I should try to compile
> > custom minimal kernel 'ident PXE' ... It is unbelievable how many
> > problems do I have ...
> > 
> > 
> > 
> > Domagoj Smolčić