On Mon, 12 Oct 1998, Shannon Hendrix wrote:
> In message <Pine.LNX.3.96.981012021726.5498C-100000@52-a-usw.rb1.blv.nwnexus.ne
> t>, Tim Smith writes:
>
> > On Mon, 12 Oct 1998, Shannon Hendrix wrote:
> > > The 53x700 eats CPU, the Adaptec seems better in this case, even
> > > the 1540B which has no CPU.
> >
> > I don't understand why a 53x700 would eat CPU. Are you sure you don't
> > mean a 53x400?
>
> Perhaps. I have an NCR 53c810 in my Alphastation that seems to
> use a lot. Hard to tell because top and vmstat don't agree on
> certain stats.
NCR53c810 devices execute SCRIPTS instructions out of main memory.
On 810 revision < 16, The SCRIPTS processor fetches instructions from the
PCI BUS using 1 DWORD PCI read at a time.
810 revision > 16 (called 810A) are able to prefetch 8 DWORDS at a time
using PCI Bursts.
Recent Symbios chips (825A, 875, 895, 896) have on-chip RAM for SCRIPTS
and so it is possible to considerably reduce PCI transactions due to
SCRIPTS fetch. The ncr53c8xx driver loads in the on-chip RAM the SCRIPTS
code that is executed for normal cases, so executed most of the time.
The only advantage of early 53C810 rev. < 16 over the 1542 is that they
use the PCI BUS, but they are also obsolete devices, in my opinion.
BTW, the new driver at ftp://ftp.tux.org/pub/roudier/896/ does not support
old 810, 815 and 825 controllers anymore. This was necessary to actually
make thinks optimal and fine for recent chips (810A,860,825A,875,895,896)
that this driver supports obviously (not only the 896).
I donnot only claim that old SCSI controllers have to go to the abattoir,
but I really did so in the new 53c8xx driver I am currently developing for
Linux.
Regards,
Gerard.
-
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/