[syslinux] pxelinux tftp timeout

[Collins, Jeremy]

> -----Original Message-----
> From: syslinux-bounces at zytor.com 
> [mailto:syslinux-bounces at zytor.com] On Behalf Of David Halik
> Sent: Thursday, May 22, 2008 5:55 AM
> To: For discussion of SYSLINUX and tftp-hpa
> Subject: Re: [syslinux] pxelinux tftp timeout
> 
> Geert Stappers wrote:
> > Op 21-05-2008 om 16:36 schreef H. Peter Anvin:
> >> David Halik wrote:
> >>> Is there a way to change the default timeout for tftp in 
> pxelinux? 
> >>> The be more specific, we use IP in HEX to network boot 
> and it takes 
> >>> a very long time for pxelinux to timeout on GUID and MAC, before 
> >>> finally getting to HEX and booting.
> >> Well, you obviously can't configure it in the configuration file, 
> >> since we're talking about how the configuration file is 
> located!  You 
> >> can send a DHCP option to give it the name of the 
> configuration file, 
> >> *however*, there is something wrong with your TFTP server.  Under 
> >> normal circumstances, there is no need to timeout here - the TFTP 
> >> server should send back an error reply, and so processing 
> the various 
> >> nonexistent filenames should take only a fraction of a second.
> > 
> 
> Hmm, well that is definitely the symptom. It sits at each 
> wrong guess for a couple of minutes before proceeding on and 
> finally getting to the correct HEX entry, so according to 
> what you're saying pxelinux is not seeing the error response 
> from the tftp server. I'll look into that some more.
> 
> > FWIW: I think it is not the TFTP server.
> > That server is probally missing name server information.
> > The long timeout time is name resolving time out.
> > 
> > 
> 
> Actually, this could be the reason! I'm curious, why is dns 
> necessary if
>   the pxeboot transactions are all IP based? DNS in our 
> network is a mysterious world because we do a lot of work 
> with private 192 IP space, virtual IP's, and load balanced 
> clusters. Effectively, there are many instances when a 
> machine's internal IP (the one DHCP gives pxelinux) does not 
> have a fully registered hostname, only it's outside vip does, 
> and this could represent a whole cluster. What I'm trying to 
> say here is that we run a very large network where most 
> servers are initially installed or brought up without 
> registered domain names because tftp is all done from point 
> to point across 192 by IP. This means DNS lookups are not 
> helpful on much of the 192 space by design. Of course, it 
> sounds like all I have to do is give the tftp server an entry 
> in /etc/hosts to test and see if it fails over correctly. 
> I'll try some of these out and see if it changes.
> 

I'd check to see if somehting on the TFTP server was doing a reverse
lookup on the hostname (ie for a log entry). From what you've said you
don't appear to have reverse DNS for the hosts and if the server is
trying to resolve client ip -> dns name for the logs that could be
slowing things up. Jeremy

> 
> > Cheers
> > Geert Stappers
> > 
> > _______________________________________________
> > SYSLINUX mailing list
> > Submissions to SYSLINUX at zytor.com
> > Unsubscribe or set options at:
> > http://www.zytor.com/mailman/listinfo/syslinux
> > Please do not send private replies to mailing list traffic.
> > 
> 
> 
> --
> ================================
> David Halik
> System Administrator
> OIT-CSS Rutgers University
> dhalik at jla.rutgers.edu
> ================================
> 
> _______________________________________________
> SYSLINUX mailing list
> Submissions to SYSLINUX at zytor.com
> Unsubscribe or set options at:
> http://www.zytor.com/mailman/listinfo/syslinux
> Please do not send private replies to mailing list traffic.
> 
>