Re: Failing to netboot

[Alexander Schreiber <als%thangorodrim.ch@localhost>]

On Tue, Apr 29, 2025 at 11:48:38AM +1000, Matthew Wallis wrote:
> 
> 
> > On 29 Apr 2025, at 03:46, mike.spooner.ux <mike.spooner.ux%gmail.com@localhost>
> > wrote:
> > 
> > I've seen this error before with multiple different O/S netboot
> > files, including even Solaris. The problem is almost always an interrupted
> > transfer (from the perspective of the OBP TFTP client, which is very
> > basic). This can sometimes be due to an incompatibility between the OBP and
> > the network infrastructure or TFTP server - especially around IP-fragmented
> > UDP packets, or shorter-than-expected (but completely valid and complete)
> > packets.
> 
> Ahh, and finally my brain catches up and calls me out for being slack.
> 
> I’ve been having network issues with the slower devices on my network. I’ll
> get NetBSD installed but then can’t pull down any packages. Transfers start
> out ok, hitting speeds of 100 - 200kB/sec then just stall.
> 
> I have a MacSE 30 as well as the IPX that exhibits the same problem. I
> recently “solved” it on the Mac by setting net.inet.tcp.*buff_max to 8MB or
> so. It would still stall, but was more likely to recover.
> 
> This also happened to packages on NFS shares. 
> 
> I should look deeper into this. My thoughts have generally been that the
> switches were not doing 10/100 very well. I tried replacing the Ubiquiti
> switch I was using with a Netgear, same problem. 

Hmm, I keep hearing anecdotes about modern (GBit) switches technically
being able to drop down all the way to 10 MBit/s, but that not working
all that well - maybe try a more ... "period correct" (e.g. tops out at
100 MBit/s) switch, which _might_ work better with 10 MBit/s clients?

> In theory this problem shouldn’t really hit the boot.net file, it’s less than
> 70k in size, and I don’t think I ever saw anything fail that soon, but it
> could be part of it.
> 
> > Getting an older TFTP server software or upgrading the OBP version can help
> > avoid this issue.
> 
> It’s on OBP 2.4 which I think is newest for these machines.
> 
> Tftp is sitting on a raspberry pi running Debian. Can try some other options.
> The contents are actually being NFS mounted from a NAS, and try moving the
> files locally too.

That NAS presumably runs _some_ version of Linux, which leaves another
trap: Does the NAS support NFSv2 and is it enabled? I don't have experience
with port-sparc netbooting, but the port-sparc64 loader only supports
NFSv2, which these days tends to be turned off by default on Linux and
is generally not longer tested by Linux distributions (IIRC Redhat went
all way to "not officially supported anymore").

Kind regards,
           Alex.
-- 
"Opportunity is missed by most people because it is dressed in overalls and
 looks like work."                                      -- Thomas A. Edison