Hi,
we intend to write large volume of data directly to disk via device
(/dev/hd1) bypassing the file system. We tested maximal transfer rates for
EIDE disks and the results are very poor for QNX4.25 as well as for Neutrino
2.1. We did not succeeded to get over 5MB/sec for block read and 3MB/sec for
block write. Parameters for used disk drivers tells about 15MB/sec or
higher. The results for Neutrino are a little worse than for QNX4.25 in PIO
modes and about 40% worse in DMA mode.
Any comments ???
Is there a possibility to run EIDE drivers for QNX4.25 in DMA mode?
Jiri Kristek
RETIA, a.s.
Jiri Kristek wrote:
Hi,
we intend to write large volume of data directly to disk via device
(/dev/hd1) bypassing the file system. We tested maximal transfer rates for
EIDE disks and the results are very poor for QNX4.25 as well as for Neutrino
2.1. We did not succeeded to get over 5MB/sec for block read and 3MB/sec for
block write. Parameters for used disk drivers tells about 15MB/sec or
higher. The results for Neutrino are a little worse than for QNX4.25 in PIO
modes and about 40% worse in DMA mode.
Any comments ???
I’ve seen reports about 10Mb/sec under Neutrino in DMA mode. You
probably have an unsupported chipset.
Is there a possibility to run EIDE drivers for QNX4.25 in DMA mode?
Not that I know of.
“Igor Kovalenko” <Igor.Kovalenko@motorola.com> wrote in message
news:39FEEEB7.7C7668EE@motorola.com…
Jiri Kristek wrote:
Hi,
we intend to write large volume of data directly to disk via device
(/dev/hd1) bypassing the file system. We tested maximal transfer rates
for
EIDE disks and the results are very poor for QNX4.25 as well as for
Neutrino
2.1. We did not succeeded to get over 5MB/sec for block read and 3MB/sec
for
block write. Parameters for used disk drivers tells about 15MB/sec or
higher. The results for Neutrino are a little worse than for QNX4.25 in
PIO
modes and about 40% worse in DMA mode.
Any comments ???
I’ve seen reports about 10Mb/sec under Neutrino in DMA mode. You
probably have an unsupported chipset.
On my laptop with a very crapy hd, I get 1.5-2 Mb/sec with QNX4 and
over 5 Mb/sec with NTO!!! And NTO uses only 25% of the cpu while
QNX4 uses over 80%
Is there a possibility to run EIDE drivers for QNX4.25 in DMA mode?
Not that I know of.
Igor Kovalenko <Igor.Kovalenko@motorola.com> píse v diskusním
pøíspìvku:39FEEEB7.7C7668EE@motorola.com…
Jiri Kristek wrote:
Hi,
we intend to write large volume of data directly to disk via device
(/dev/hd1) bypassing the file system. We tested maximal transfer rates
for
EIDE disks and the results are very poor for QNX4.25 as well as for
Neutrino
2.1. We did not succeeded to get over 5MB/sec for block read and 3MB/sec
for
block write. Parameters for used disk drivers tells about 15MB/sec or
higher. The results for Neutrino are a little worse than for QNX4.25 in
PIO
modes and about 40% worse in DMA mode.
Any comments ???
I’ve seen reports about 10Mb/sec under Neutrino in DMA mode. You
probably have an unsupported chipset.
Is there a possibility to run EIDE drivers for QNX4.25 in DMA mode?
Not that I know of.
Is it possible to ask somehow to QNX if there are plans to provide DMA and
UDMA drivers for EIDE disks ?
Previously, Jiri Kristek wrote in comp.os.qnx:
we intend to write large volume of data directly to disk via device
(/dev/hd1) bypassing the file system. We tested maximal transfer rates for
EIDE disks and the results are very poor for QNX4.25 as well as for Neutrino
2.1. We did not succeeded to get over 5MB/sec for block read and 3MB/sec for
block write. Parameters for used disk drivers tells about 15MB/sec or
higher. The results for Neutrino are a little worse than for QNX4.25 in PIO
modes and about 40% worse in DMA mode.
Any comments ???
Is there a possibility to run EIDE drivers for QNX4.25 in DMA mode?
To bypass Fsys entirely, you might try using Blkfsys. The reason to
try this is that Fsys caches direct i/o through /dev/hd#, whereas
Blkfsys does not. This might create a problem accessing /dev/hd0
as a file system. I once solved this problem in a kludge manner by
creating a psuedo driver that read /dev/hd0 directly. This driver
was owned by Fsys, which was therefore requesting blocks through
Blkfsys. There is some overhead in this process, but it allows
direct I/O of /dev/hd1 through Blkfsys.
You also could try to write your own DMA mode driver for EIDE.
Mitchell Schoenbrun --------- maschoen@pobox.com