Syslinux - Jan 2005 - Slow performance booting USB device with 3.xx version

With the new 3.01 and 3.07 version I found a big slowdown booting from
USB devices.

On my PC it takes more than 2 minutes to load the kernel and the
initrd filesystem. Compared with the previous behaviour of syslinux
2.13 with takes only 10 seconds or little more.

The BIOS boots in FDD mode (DriveNumber variable is 0). I've also
tried to disable the EDD support in the source. But the boot process
takes always a long time.

I've also a PC which boots in HDD mode (DriveNumber has the highest
bit set) and it is also slower, approx 20 seconds compared to the
previous 10.

I don't understand where this slowdown come from. Which changed
exactly from the 2.13 version ? Peter, do you have some hints for my
tests ?

I'm very interested in the new FAT32 support, but I don't like to have
regression from the previous version.

-- 
Andrea Mazzoleni
935A 2D3C 5C70 BCD6 CB0C  ED89 7C19 4321 6340 3F6D

Andrea Mazzoleni wrote:
> With the new 3.01 and 3.07 version I found a big slowdown booting from
> USB devices.
> 
> On my PC it takes more than 2 minutes to load the kernel and the
> initrd filesystem. Compared with the previous behaviour of syslinux
> 2.13 with takes only 10 seconds or little more.
> 
> The BIOS boots in FDD mode (DriveNumber variable is 0). I've also
> tried to disable the EDD support in the source. But the boot process
> takes always a long time.
> 
> I've also a PC which boots in HDD mode (DriveNumber has the highest
> bit set) and it is also slower, approx 20 seconds compared to the
> previous 10.
> 
> I don't understand where this slowdown come from. Which changed
> exactly from the 2.13 version ? Peter, do you have some hints for my
> tests ?
> 
> I'm very interested in the new FAT32 support, but I don't like to
have
> regression from the previous version.
> 
The two differences are EDD support, and the fact that the FAT isn't all 
sucked in before the command prompt is shown.  Instead there is a 64K 
cache used for the FAT and directory information.

This should pretty much amount to an extra access every 256 sectors (for 
FAT16), but I have a very hard time imagining the construction of a USB 
BIOS which would suffer that greatly.

The fact that the BIOS boots in floppy mode is an indication that 
something is very wrong to begin with.  Have you managed to get it into 
HDD mode, or at least zip mode?  If you're in floppy mode, what's the 
geometry?

It's possible that you're regressing into sector-at-a-time mode, but 
that's very hard for me to know without access to the particular system 
that you're on.

	-hpa

Andrea Mazzoleni wrote:
> 
> I've also a PC which boots in HDD mode (DriveNumber has the highest
> bit set) and it is also slower, approx 20 seconds compared to the
> previous 10.
> 
One important thing about benchmarking boot time is to make sure to set 
"prompt 0" and time from reset (or any other fixed point in the bootup
sequence) until "Ready".  Any other timing is subject to being shifted
around, and there is definitely time having been shifted from SYSLINUX 
startup to kernel load time.

	-hpa