ethernet performance on PPC405

Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut, PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip devices.


verbose Disable emission control of message output. ← does
this mean: be verbose?

Kind Regards,
Werner Benz

Hi Werner,

We’re facing exactly the same performance problem and just like you we
didn’t notice any improvements after switching to release 6.2.1. We’re using
Embedded Planet’s EP405PC Board, 200MHz, 64MB RAM. The highest FTP transfer
rate that we could reach was only about 0.7-0.8 MByte/sec - way below
expected.

I was also adviced by QNX support people to add two more options: “-c1” and
“cache=1”, although it didn’t help much as I mentioned. That’s the command
line use:

io-net -c1 -d ppc405 … -p tcpip cache=1 &


It’s been 2 months since this message was posted (no followups/responses).
Does it mean the problem has been resolved?

Jaroslav

==========================
Jaroslav Ovtsyn
Embedded Software Designer
jrslv@rogers.com
613-836-8813
Ottawa, ON, Canada


“Werner Benz” <werner.benz@ndt-ag.de> wrote in message
news:b08gtj$la7$1@inn.qnx.com

Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a
better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut,
PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip devices.
.
.
verbose Disable emission control of message output. ← does
this mean: be verbose?

Kind Regards,
Werner Benz

Jaroslav Ovtsyn <jrslv@rogers.com> wrote:

Hi Werner,

We’re facing exactly the same performance problem and just like you we
didn’t notice any improvements after switching to release 6.2.1. We’re using
Embedded Planet’s EP405PC Board, 200MHz, 64MB RAM. The highest FTP transfer
rate that we could reach was only about 0.7-0.8 MByte/sec - way below
expected.

Werners original post was posted to two groups, and the thread continued
in the other group…

In any case, the throughput you’re seeing is way below what you should
be getting. Can you post the output of “nicinfo” after running some
ftp tests? Also, what is the Rev. number written on the 405 chip
that you are using?

Regard,
Dave

I was also adviced by QNX support people to add two more options: “-c1” and
“cache=1”, although it didn’t help much as I mentioned. That’s the command
line use:

io-net -c1 -d ppc405 … -p tcpip cache=1 &



It’s been 2 months since this message was posted (no followups/responses).
Does it mean the problem has been resolved?

Jaroslav

==========================
Jaroslav Ovtsyn
Embedded Software Designer
jrslv@rogers.com
613-836-8813
Ottawa, ON, Canada



“Werner Benz” <> werner.benz@ndt-ag.de> > wrote in message
news:b08gtj$la7$> 1@inn.qnx.com> …
Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a
better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut,
PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip devices.
.
.
verbose Disable emission control of message output. ← does
this mean: be verbose?

Kind Regards,
Werner Benz

Hi,
yes my original post was also in the 6.2.1 beta group which is closed now.
Because I hoped for feedback from other PPC users I posted it also in an
public group, nice to see any response for that :slight_smile:
The problem is still there, there was a increasing performance with a driver
we got from QNX support and now we reach about 2 -3 MByte/sec with qnet
between nodes, tested with netio and own io test programs. FTP from a
windows client works up to 1 MByte/sec, performance strongly depends on the
size of the transferred blocks. Compared to Linux on the same hardware this
are poor 30% from that what is possible on that hardware.
German support give us a further outlook to version QNX 6.3, which will
increase the performance of several io-drivers and the contiguous cache
handling. So if we decide to stay with QNX we will have to work resigned
with the poor performance with the glimmer of hope the next version of
QNX6.3.
Regard,
Werner


“David Donohoe” <ddonohoe@qnx.com> wrote in message
news:b5dqgv$1lu$1@nntp.qnx.com

Jaroslav Ovtsyn <> jrslv@rogers.com> > wrote:
Hi Werner,

We’re facing exactly the same performance problem and just like you we
didn’t notice any improvements after switching to release 6.2.1. We’re
using
Embedded Planet’s EP405PC Board, 200MHz, 64MB RAM. The highest FTP
transfer
rate that we could reach was only about 0.7-0.8 MByte/sec - way below
expected.

Werners original post was posted to two groups, and the thread continued
in the other group…

In any case, the throughput you’re seeing is way below what you should
be getting. Can you post the output of “nicinfo” after running some
ftp tests? Also, what is the Rev. number written on the 405 chip
that you are using?

Regard,
Dave

I was also adviced by QNX support people to add two more options: “-c1”
and
“cache=1”, although it didn’t help much as I mentioned. That’s the
command
line use:

io-net -c1 -d ppc405 … -p tcpip cache=1 &


It’s been 2 months since this message was posted (no
followups/responses).
Does it mean the problem has been resolved?

Jaroslav

==========================
Jaroslav Ovtsyn
Embedded Software Designer
jrslv@rogers.com
613-836-8813
Ottawa, ON, Canada


“Werner Benz” <> werner.benz@ndt-ag.de> > wrote in message
news:b08gtj$la7$> 1@inn.qnx.com> …
Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a
better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut,
PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with
the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect
data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with
any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip
devices.
.
.
verbose Disable emission control of message output. ←
does
this mean: be verbose?

Kind Regards,
Werner Benz

Hi David,

We’re facing exactly the same performance problem and just like you we
didn’t notice any improvements after switching to release 6.2.1. We’re
using
Embedded Planet’s EP405PC Board, 200MHz, 64MB RAM. The highest FTP
transfer
rate that we could reach was only about 0.7-0.8 MByte/sec - way below
expected.

I think this needs to be corrected:
– 0.8 MByte/sec is reached when a file is transferred into the target’s RAM
disk using FTP.
– around 2MByte/sec is reached when the data are dropped by the target
immediately after receiving.

In my test setup - the target is on the receiving side of a TCP/IP
connection, while the host transmits data non-stop - the transfer rate drops
significantly every few seconds. I attribute this behavior to a TCP flow
control mechanism; they probably have some sort of a back-off algorithm to
cope with network congestions.

I tried to increase TCP TX/RX buffers using setsockopt(), but the system
caps them at around 20K. Any idea how could I do that ?

Werners original post was posted to two groups, and the thread continued
in the other group…

In any case, the throughput you’re seeing is way below what you should
be getting. Can you post the output of “nicinfo” after running some
ftp tests? Also, what is the Rev. number written on the 405 chip
that you are using?

This is what’s written on the chip:
PowerPC 405GP
03BM
1E10A156PB
IBM25PPC405GP-3DE200C

And this is what I get from “nicinfo” (nothing suspicios, huh?) :

PPC405 on-chip EMAC Ethernet Controller
Physical Node ID … 0010EC 000000
Current Physical Node ID … 0010EC 000000
Media Rate … 100.00 Mb/s full-duplex UTP
MTU … 1514
Lan … 0
I/O Port Range … 0xEF600800 → 0xEF6008FF
Hardware Interrupt … 0xC
Promiscuous … Disabled
Multicast … Enabled
Total Packets Txd OK … 35828
Total Packets Txd Bad … 0
Total Packets Rxd OK … 160037
Total Rx Errors … 0
Total Bytes Txd … 2364640
Total Bytes Rxd … 237865911
Tx Collision Errors … 0
Tx Collisions Errors (aborted) … 0
Carrier Sense Lost on Tx … 3
FIFO Underruns During Tx … 0
Tx deferred … 0
Out of Window Collisions … 0
FIFO Overruns During Rx … 0
Alignment errors … 0
CRC errors … 0

Regards,
Jaroslav


Regard,
Dave

I was also adviced by QNX support people to add two more options: “-c1”
and
“cache=1”, although it didn’t help much as I mentioned. That’s the
command
line use:

io-net -c1 -d ppc405 … -p tcpip cache=1 &


It’s been 2 months since this message was posted (no
followups/responses).
Does it mean the problem has been resolved?

Jaroslav

==========================
Jaroslav Ovtsyn
Embedded Software Designer
jrslv@rogers.com
613-836-8813
Ottawa, ON, Canada


“Werner Benz” <> werner.benz@ndt-ag.de> > wrote in message
news:b08gtj$la7$> 1@inn.qnx.com> …
Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a
better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut,
PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with
the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect
data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with
any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip
devices.
.
.
verbose Disable emission control of message output. ←
does
this mean: be verbose?

Kind Regards,
Werner Benz

Jaroslav Ovtsyn <jrslv@rogers.com> wrote:

This is a multi-part message in MIME format.

------=_NextPart_000_0015_01C2F45F.3747D480
Content-Type: text/plain;
charset=“iso-8859-1”
Content-Transfer-Encoding: 7bit

Hi David,

We’re facing exactly the same performance problem and just like you we
didn’t notice any improvements after switching to release 6.2.1. We’re
using
Embedded Planet’s EP405PC Board, 200MHz, 64MB RAM. The highest FTP
transfer
rate that we could reach was only about 0.7-0.8 MByte/sec - way below
expected.

I think this needs to be corrected:
– 0.8 MByte/sec is reached when a file is transferred into the target’s RAM
disk using FTP.
– around 2MByte/sec is reached when the data are dropped by the target
immediately after receiving.

In my test setup - the target is on the receiving side of a TCP/IP
connection, while the host transmits data non-stop - the transfer rate drops
significantly every few seconds. I attribute this behavior to a TCP flow
control mechanism; they probably have some sort of a back-off algorithm to
cope with network congestions.

I tried to increase TCP TX/RX buffers using setsockopt(), but the system
caps them at around 20K. Any idea how could I do that ?

It sounds like packets are being dropped and retransmitted. Try
tuning the recvspace parameter, e.g.

sysctl -w net.inet.tcp.recvspace=8192

You might get some more clues as to what’s going on by running
netstat -ptcp and netstat -pip.

Werners original post was posted to two groups, and the thread continued
in the other group…

In any case, the throughput you’re seeing is way below what you should
be getting. Can you post the output of “nicinfo” after running some
ftp tests? Also, what is the Rev. number written on the 405 chip
that you are using?

This is what’s written on the chip:
PowerPC 405GP
03BM
1E10A156PB
IBM25PPC405GP-3DE200C

And this is what I get from “nicinfo” (nothing suspicios, huh?) :

Everything looks normal here.

PPC405 on-chip EMAC Ethernet Controller
Physical Node ID … 0010EC 000000
Current Physical Node ID … 0010EC 000000
Media Rate … 100.00 Mb/s full-duplex UTP
MTU … 1514
Lan … 0
I/O Port Range … 0xEF600800 → 0xEF6008FF
Hardware Interrupt … 0xC
Promiscuous … Disabled
Multicast … Enabled
Total Packets Txd OK … 35828
Total Packets Txd Bad … 0
Total Packets Rxd OK … 160037
Total Rx Errors … 0
Total Bytes Txd … 2364640
Total Bytes Rxd … 237865911
Tx Collision Errors … 0
Tx Collisions Errors (aborted) … 0
Carrier Sense Lost on Tx … 3
FIFO Underruns During Tx … 0
Tx deferred … 0
Out of Window Collisions … 0
FIFO Overruns During Rx … 0
Alignment errors … 0
CRC errors … 0

Regards,
Jaroslav



Regard,
Dave

I was also adviced by QNX support people to add two more options: “-c1”
and
“cache=1”, although it didn’t help much as I mentioned. That’s the
command
line use:

io-net -c1 -d ppc405 … -p tcpip cache=1 &


It’s been 2 months since this message was posted (no
followups/responses).
Does it mean the problem has been resolved?

Jaroslav

==========================
Jaroslav Ovtsyn
Embedded Software Designer
jrslv@rogers.com
613-836-8813
Ottawa, ON, Canada


“Werner Benz” <> werner.benz@ndt-ag.de> > wrote in message
news:b08gtj$la7$> 1@inn.qnx.com> …
Hello PowerPC users,

one intention to participate as beta-tester for QNX6.2.1 is to get a
better,
re-designed devn-ppc405.so, announced from german QNX-Support group.
Now we built the BSP’s, images for different PPC405 boards (Walnut,
PIP405,
MIP405 from MPL AG and proprietary hardware with PPC405), running with
the
newest release, but there is no change in ethernet performance. The
bandwidth that we could reach via QNET is about 2MByte/sec, we expect
data
rates above 6MBytes/sec (7.5MByte/sec measured with Linux on PPC405).
Does anyone test the ethernet performance on other PowerPC boards with
any
QNX release? Which data rates could you reach?

One point to the devn-ppc405.so:
calling the usage info tells following:

Syntax:
io-net -d ppc405 [option[,option …]] … &

Options
deviceindex Only attach to this instance of the on-chip
devices.
.
.
verbose Disable emission control of message output. ←
does
this mean: be verbose?

Kind Regards,
Werner Benz




------=_NextPart_000_0015_01C2F45F.3747D480
Content-Type: text/x-vcard;
name=“Jaroslav Ovtsyn.vcf”
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename=“Jaroslav Ovtsyn.vcf”

BEGIN:VCARD
VERSION:2.1
N:Ovtsyn;Jaroslav
FN:Jaroslav Ovtsyn
TITLE:Embedded Software Specialist
TEL;HOME;VOICE:(613) 836-8813
TEL;HOME;FAX:(613) 836-8813
ADR;HOME:;;33 Denham Way;Stittsville;ON;K2S 1H5;Canada
LABEL;HOME;ENCODING=3DQUOTED-PRINTABLE:33 Denham =
Way=3D0D=3D0AStittsville, ON K2S 1H5=3D0D=3D0ACanada
EMAIL;PREF;INTERNET:> jrslv@rogers.com
REV:20030327T174850Z
END:VCARD

------=_NextPart_000_0015_01C2F45F.3747D480–