On Tue, 5 Oct 1999, Kurt Garloff wrote:
> On Sun, Oct 03, 1999 at 04:14:59PM +0200, Gerard Roudier wrote:
> > In fact the Tekram DC-3x5 series of PCI-SCSI adapters donnot
> > use LSI/SYMBIOS 53C8XX chips, but a Tekram proprietary chip,
> > they named S1040.
> >
> > This chip is very different from the 53C8XX family and needs
> > different drivers. Tekram provides some drivers for their
> > DC-3x5 controllers from their ftp site.
> >
> > People who want Tekram boards that use 53C8XX chips must not
> > purchase these new boards, but order one of the following:
> > DC-310, DC-390-U, DC-390-F, DC-390-U2B, DC-390-U2W that are
> > excellent products.
>
> True.
I didn't expect to be wrong here. :)
> > I have looked into Tekram driver sources for the S1040 chip, and
> > it seems that this chip does not implement a hardware phase engine.
> >
> > That means that all phase changes must be handled from the C
> > code. May-be, it is their driver that is not optimal, may-be
> > the S1040 chip is actually designed so.
> > Result is:
> > - At least 5 interrupts per SCSI transfer (instead of about 1 with
> > 53C8XX family and aic7xxx family that implement a hardware phase
> > engine)
> > - Far more IOs from the C code per SCSI transfer.
> > - More CPU load per SCSI transfer.
>
> True. It does not have something like a script processor or similar. It does
Neither there, btw. :)
> invoke an interrupt every time the SCSI phase changes.
> The design is similar to the AM53C974 chip plus a few extras like Scatter
> Gather hardware support.
>
> BTW: You can find a driver on ftp:://ftp.suse.com/pub/people/garloff/linux/dc395/
Seems Tekram also provides drivers for Linux and FreeBSD.
> > On the comparison chart which is available at their Web site, they
> > announce for the DC-3X5 family about the same performance as the
> > boards using a 53C8XX chip. Some of the number are a bit better for
> > the DC-3X5 family.
>
> If you use an OS that does have to do other things while waiting for I/O, it
> does not make a difference. Even on Linux, the difference to those adapters
> implementing script processors is quite small, unless your I/O load gets very
> high.
Using SCSI with low IO load is not that interesting, given the extra cost.
In fact the difference may be large as well as the interrupt latency (PCI)
and load, since this chip also requires to do some extra IO to PCI from
the C code. Such a design for a modern PCI-SCSI adapter is a shame.
Better to use IDE/ATAPI than such ultra-poor SCSI stuff as the S1040,
IMNSO.
Regards,
G�rard.
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
