[syslinux] syslinux tool for Win2k and 'safeboot' option

Patrick J. LoPresti patl at curl.com
Sat Oct 26 15:05:54 PDT 2002


"H. Peter Anvin" <hpa at zytor.com> writes:

> Patrick J. LoPresti wrote:
> > Wouldn't this be fundamentally flaky, since TFTP is not a reliable
> > protocol?  That is, the client could request the file, then the server
> > would send it, then the client might request it again because a packet
> > got dropped?
> >
> 
> ALL protocols are fundamentally unreliable, including TCP.  It's just
> a matter of the unreliability.

I suppose I should have said "designed for reliability" instead of
"reliable".

> However, you're wrong about the level of unreliability of TFTP.

I phrased my message as a question, because I had not yet read the
TFTP spec.  Now I have, and thank you for the encouragement.

> TFTP contains a packet retransmit mechanism, and a complete TFTP
> transfer failure is rare.  Besides, you could configure your server
> to only change state after a successful transfer.

You are right, the only realistic failure mode is for the final ACK to
get dropped, meaning the server will think the transfer failed when it
actually succeeded.  This is not a bad failure mode.

> Sure, but how do you get the client back to a responsive state?
> I am questioning the value of this feature because it seems to assume
> that a remote way to reset the system exists.

Somebody is bound to power-cycle the system eventually :-).

More seriously, you could use "watchdog timer" hardware to bounce the
system in general when it becomes unresponsive.

Still, it would be nice if there were someplace to store data between
reboots independent of the filesystem.  My interest in this is
unattended Windows installation, where the initial disk partitioning
(from DOS) requires rebooting the machine.  I would like my
installation process to be able to pick up where it left off, which is
tricky when you are editing the partition table and rebooting...

 - Pat



More information about the Syslinux mailing list