Re: Ext3 strangeness data loss

["Stephen C. Tweedie" <sct redhat com>]

Hi,

On Mon, 2003-02-03 at 22:11, Matthias Andree wrote:

> Looking at hardware and this issue, the German c't magazine recently
> reported that ready-to-run computers were shipped with ATA cables that
> exceeded the maximum specified 45 cm (18 in) length. These bring the
> danger zone much nearer than desired...

Yes, but not as much as you'd imagine.  UDMA transfers are protected by
a CRC between drive and controller, so if you have cable problems at the
highest speeds, it is supposed to retry and then back down to a slower
speed, at which the cable length requirements are relaxed.

It's only if you end up backing down from UDMA entirely (to [M]DMA or
PIO) that you lose the CRC.  At that point,
slightly-too-long-for-UDMA100 cables really don't matter, but other
cable problems such as dirty connectors can make a big difference.

At the higher speeds, I've actually found that it's the motherboard-side
transfers which seem to be most impacted, with various chipset bugs
causing things like a single lost line of DMA between the controller and
main memory.

Cheers,
 Stephen