> On Sun, 3 Oct 1999, Matthew Jacob wrote:
>
> > 126 for private arbitrated loop, massively larger for switch fabrics.
>
> BTW, arbitrated loop absolutely sucks if you're in a hi-av clustered
> environment. By definition, every host's card has to pass along packets
> not meant for itself. If you reboot one of the nodes while I/O is
> happening, that I/O gets clobbered with *NO* indication that it hit the
> bit bucket. Proceed slowly to timeout, do not collect the buffer, do not
> pass go.
Yes, I've been struggling with this for the whole year. It isn't entirely
as bad as this, but close to it. You typically don't wait for a
timeout because you get a LIP or a Loop Reset close to the reboot. You
also don't necessarily have to always worry terribly much about the
packets just not getting passed through because the GBICs fail into bypass
mode.
The trouble here is that arbitrated loop and the hubs for it is affordable
while switch fabrics are still too pricey. You can *almost* buy arb loop
cards off the shelf and expect them to work like fat scsi busses.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/