EPOX nVIDIA nForce3 250Gb Manuel d'utilisateur Page 4

  • Télécharger
  • Ajouter à mon manuel
  • Imprimer
  • Page
    / 11
  • Table des matières
  • MARQUE LIVRES
  • Noté. / 5. Basé sur avis des utilisateurs
Vue de la page 3
auxiliary connector (short VSI cable coming out of the PC), instead of the copy of
input which is regularly there. The data output (be it TVG data or duplicated data
from the primary input) of the auxiliary VSI-H connector is unaffected by VSIB
board operation, that is, regardless whether the VSIB board is stopped, waiting, or
running. Please note that the TVG VSI data stream being outputted in this case is
indeed “overclocked” at 50 MHz.
3.2 VSIB Board Operation
The VSIB board has a very simple operating model, one settable 32-bit word of set-
tings and an internal state, one of three possible states: stopped, waiting, running.
Regularly, the board is in the “stopped” state where it resets all its internal logic
to a known state. This is reflected in the back panel LED number 1 (and also on-
board D4). At any time the driver software can reset VSIB back to this state by just
writing into the 32-bit settings/mode word.
The board is started by writing another bit combination (the start bit together with
all the settings required) into the 32-bit settings/mode word. The on-board logic
lights up the panel LED number 2 (and also on-board D5) and starts “hunting” for
the next occurrence of VSI 1pps pulse, keeping the logic in this “waiting” state.
The first data word being clocked in with the VSI 1pps signal asserted is the first
word being captured and this triggers the board into the “running” state and lights
up the panel LED number 3 (and also on-board D6). The board starts pushing in-
coming VSI words into its on-board 4 kB FIFO and from there into the PCI bus us-
ing DMA. The words land into a Linux “bigphysarea” main memory buffer where
the ’vsib.o’ driver finds them and presents them to the user mode programs via
regular ’read()’ calls. The main memory buffer is organized as a “scatter-gather
circular buffer and it is allocated when the ’vsib.o’ device driver is loaded with the
argument ’bigbufsize’:
insmod /home/amn/proj/vsib/vsib.o bigbufsize=144000000
The above allocates 144000000 bytes for the ring buffer; generally at 512 Mbps less
than 10 Mbytes is sufficient. The maximum consumption of this is being tracked
with kernel log messages which appear in Linux virtual console number 1 (Alt-F1)
and are also recorded in /var/log/kern.log:
vsib: big secondary ring buffer filled to 3456000 bytes
Should this reach a value very close to 144000000 it very probably indicates that the
disk subsystem is not keeping up with the VSIB PCI DMA. A set of three Maxtor
200GB PATA disks can sustain a little over 700 Mbps at the beginning of disks but
at the inner tracks, during the last 2% of disk capacity the sustained performance
is very close to 512 Mbps. Another typical reason for sudden disk slowness is the
attempt to overwrite large old files with new files with the same file names: delet-
ing large files “on-the-fly” takes a relatively long time and the disk write process
is left behind the PCI DMA rate.
The panel LED number 4 (and also on-board D7) indicates that the on-board PCI
FIFO has overflowed. This could happen if the PCI bus is unable to DMA data
quickly enough from VSIB into main memory. nVidia nForce chipsets are quite
good at this, though, and this problem starts to appear at closer to 800 Mbps.
4
Vue de la page 3
1 2 3 4 5 6 7 8 9 10 11

Commentaires sur ces manuels

Pas de commentaire