This message is from the T13 list server.

> From: Hale Landis [mailto:[EMAIL PROTECTED]]
> I just wanted to add here that I will stop posting
> my replys ... here

I'll miss the support & help I get offline & online,
though certainly I'll be pleased if I start feeling
heard as much offline as I do online.

I'd welcome an offline person-to-person reply to me
if anyone does object to us continuing my education
here?  I haven't heard any complaints since imposing
the flame limit of one post/ topic/ author/ day.

> From: Elliott, Robert (Server Storage) [[EMAIL PROTECTED]]
...
> The SCSI architecture model requires byte
> granularity for data transfers by all SCSI
> transport protocols (see sam3r04 section 5.4.3
> - Data transfer protocol services).

Good to see this real need formally acknowledged.

> > I don't see T13 doing anything to change the
> > ATA/ATAPI parallel interface data transfer
> > protocols.
>
> It's obviously survived without byte granularity so
> far.

I beg to differ.  Atapi Pio has byte granularity on
offer, for anyone who wants it.  Only Dma is broken.

> Other serial SCSI protocols all have ways to
> indicate odd lengths (often with a "number of pad
> bytes" field).

Thanks for saying, I didn't know that.

> From: Hale Landis [mailto:[EMAIL PROTECTED]]
> Pat, do you think there can be more than one pad
> byte in any PIO or DMA data transfer?

I think Yes, but I'm not convinced I understand the
question properly.

1) Of course there "can be".  Ops can be designed to
always includes lots of pad bytes no matter the
transport, much like T13's Atapi Command Out always
rounds up the Cdb Length to 12 or 16.

But in context here, I think we mean to be discussing
how many pad bytes does Ata/pi Pio/Dma data transport
force the host & the device to add, beyond what they
would agree to copy over a byte-wise synchronous
connection, like the full handshake of classic
so-called "asynchronous" 8-bit Scsi.

2) Agreed, whenever the host & the device do agree
out-of-band in advance how many data bytes to copy
which way, then Ata/pi Pio/Dma never forces them to
add more than one pad byte.

However, such an agreement by definition cannot be in
place if the command in question is a new standard or
an old proprietary command.

3) Furthermore, even with a nominal agreement, as Dma
burst rates increase past SwDma, when the end of the
bus clocking the data asks to copy more than the
other end, we can end up with more than one pad byte.

For example, I'm told many devices stop asking to
copy read/write block data sometime near the first
block in error.  Any time that does happen, the
device has stopped the host short.

> ...

Yours at least til Thursday, Pat LaVarre

Reply via email to