Speed of plb_temac 3.00 on ML403

Hi all,

I have to interface to a camera with the GigE Vision protocol.

For that I set up a design on ML403 with PPC, Temac and sgDMA. I use MontaVista 2.4.20 Linux with the BSP from Xilinx EDK 8.2. The camera vendor supplied a library for GigE Vision, which works under Linux.
The results of some tests showed that I have to insert waiting time between frames sent by the camera, otherwise the lib signals errors.
This leeds to only 1/10 of the needed transfer rate. 

My question is now: Has anybody deeper knowledge how ethernet and sgDMA works? How deep is the PPC involved in the data transfer? Or does the Temac-core handle the datatransfer to DDR-memory autonomous?

I learnd from the camera vendor, that on PCs with special Intel ethernet chips, it works outonomous and so the high transfer rate can be accomplished. 

I'm very interested to get in contact with people who have to interface with a GigE Vision camera.

Also interesting is if anybody benchmarked the gigabit ethernet on the ML403 hardware. How fast is the gigabit interface realy?

Thomas


-- 
Thomas Denzinger
LesaMetric GmbH 
Hauptstrasse 46
35649 Bischoffen

Tel.: 06444/931928
Fax : 06444/931912

Dear Thomas,
Now I am also fighting against improving the performance of the TEMAC on ml403 board. But the difference is I work on the old plb_temac and hard_temac cores and my linux is 2.6.

I used some little programs written by myself to test the throughput of the network between a pc and my board. (I would like to say, I am not an experienced programmer so I cannot make sure my programs have no problem, although they are simple. :-) ) According to my result, the largest throughput between my board and a pc is around 4MB/s, for a round-trip. However when those programs applied on two servers interconnected by gigabit ethernet, the largest throughput is around 12MB/s, three times than my board.

I have not used the popular software, such as Netperf, to measure my little network.

I also noticed that when transfering the data, the CPU utilization is very high(almost 100%). I am not sure if this is the reason. I think because there is SGDMA who can deal with data transfer, the CPU utilization should not be so high.

I am also waiting for some suggestions from some experts who are more experienced in such field.

Did you ever measure the throughput of your network? If possible, let's share some experience. :)

BR
Ming

From: Thomas Denzinger <redacted>
To: <redacted>
Subject: Speed of plb_temac 3.00 on ML403
Date: Tue, 05 Dec 2006 17:18:07 +0100

Hi all,

I have to interface to a camera with the GigE Vision protocol.

For that I set up a design on ML403 with PPC, Temac and sgDMA. I use MontaVista 2.4.20 Linux with the BSP from Xilinx EDK 8.2. The camera vendor supplied a library for GigE Vision, which works under Linux.
The results of some tests showed that I have to insert waiting time between frames sent by the camera, otherwise the lib signals errors.
This leeds to only 1/10 of the needed transfer rate.

My question is now: Has anybody deeper knowledge how ethernet and sgDMA works? How deep is the PPC involved in the data transfer? Or does the Temac-core handle the datatransfer to DDR-memory autonomous?

I learnd from the camera vendor, that on PCs with special Intel ethernet chips, it works outonomous and so the high transfer rate can be accomplished.

I'm very interested to get in contact with people who have to interface with a GigE Vision camera.

Also interesting is if anybody benchmarked the gigabit ethernet on the ML403 hardware. How fast is the gigabit interface realy?

Thomas


--
Thomas Denzinger
LesaMetric GmbH
Hauptstrasse 46
35649 Bischoffen

Tel.: 06444/931928
Fax : 06444/931912
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded

_________________________________________________________________
免费下载 MSN Explorer:   http://explorer.msn.com/lccn 

My question is now: Has anybody deeper knowledge how ethernet and sgDMA
works? How deep is the PPC involved in the data transfer? Or does the
Temac-core handle the datatransfer to DDR-memory autonomous?

Thomas,

If you cut & pasted directly from my design you may be running without
DMA, which in turn implies running without checksum offload and DRE.
The plb_temac shrinks to about half it's size this way, but if you're
performance bound you probably want to turn DMA back on in your mhs
file:

 PARAMETER C_DMA_TYPE = 3
 PARAMETER C_INCLUDE_RX_CSUM = 1
 PARAMETER C_INCLUDE_TX_CSUM = 1
 PARAMETER C_RX_DRE_TYPE = 1
 PARAMETER C_TX_DRE_TYPE = 1
 PARAMETER C_RXFIFO_DEPTH = 32768

You'll have to regenerate the xparameters file too if you make these
changes (in xps: Software -> Generate Libraries and BSPs).

There may also be issues with the IP stack in the 2.4 linux kernels.
If you have the option, an experiment with at 2.6 stack would be
ammusing.

-michael