[ECOS] Ethernet driver /MPC860 FADS
Claus Jarnik
jarnik@esk.fhg.de
Mon Dec 20 02:16:00 GMT 1999
Hello,
let me first introduce myself. My name is Claus Jarnik and I'm a student
of the Fachhochschule Munich/University of Applied Sciences.
Actually I do my diploma for the Fraunhofer ESK (research institute).
My business is the development of an ethernet driver under eCos for the
Motorola MPC860 (evaluation board: FADS).
Also I want to write something about eCos (if I get enough information
about the purpose of eCos, known commercial Applications and products ÃÂ
)
If you have any interest in my driver (end version is planned to be in
march 2000) I will make the code available to the eCos community.
Right now I might need your help because I got some problems with the
data rate of my driver.
It would be a great help for me if you have any good idea to solve the
problem.
I use the eCos cyg_io_read and cyg_io_write function for
receive/transmit data.
(cyg_io_write(cyg_io_handle_t handle, const void *buf, cyg_uint32 *len)
In these functions I copy the data from/to buf with the desired length
len.
This copying (I use memcpy) costs a lot of performance.
For a simple performance test I transmit 50000 ethernet frames (1518
bytes) from a Linux PC to the FADS board.
This are my data rates for receiving:
Action | average data rate| busy errors
-------------------------------------------------------------------------------------
Receive and check | 1.4 Mbit/s | 49997
(check if every single byte is ok) | |
Receive without check | 2.8 Mbit/s | 50001
Without memcpy (no copying of data)| 7.1 Mbit/s | 3400
Remarks: 64 receive buffer (1520 bytes per buffer), one user task, BSY
INT deactivated, non-blocking read modus
The 7.02 Mbit/s rate would be the best theoretical average data rate
with my driver.
The copying of the received data costs nearly 60% of the theoretical
data rate.
Does eCos support pointer handling with a memory pool (only to handle
pointers to data, not copying the data) because the copying of data from
layer to layer will decrease a lot of performance?
I think a pointer handling from layer to layer would be more effective.
Thank you very much
--
kind regards
Fraunhofer Einrichtung für
Systeme der Kommunikationstechnik
mailto:jarnik@esk.fhg.de
Claus Jarnik HansastraÃÂe 32
D-80686 München
Telefon: +49(0)89/547088-40
Telefax: +49(0)89/547088-20
More information about the Ecos-discuss
mailing list