[prev in list] [next in list] [prev in thread] [next in thread] 

List:       uclinux-dev
Subject:    [uClinux-dev] Getting Ethernet overrun - Need Help
From:       Nagaraj Kanniah <nagik_2k3 () yahoo ! com>
Date:       2003-05-31 5:47:53
[Download RAW message or body]

[Attachment #2 (multipart/alternative)]


Hi,
 
I am sorry, my previous mail got inadvertantly posted before I completed it.
 
We've put uclinux onto an embedded system that is based on the
Samsung S3c4510B processor. This system has two ethernet
ports - one is provided by the processor (used for LAN port) and the 
other is using a ethernet chip ( AX88796 - ASIX used for WAN port). 
The LAN chip is connected to the "external interrupt" line of the processor.
 
Everything as far as the system is concerned is ok. However, using this
box as a gateway between two networks (n1 and n2) we found
that there were some throughput problems. From a host H1(on N1)
we did a FTP put of a file to host h2(on N2). The throughput observed
was 30 KB/s. Instead of a FTP put we did a get and observed
a throughput of 550 KB/s. We were trying to debug this asymmetric
throughput problem. 
 
We observed that when doing a FTP put, we observed the output
of the "ifconfig" after a put and found around 15% of total packets
are "overrun" on the LAN port of the gateway (this is the internal
ethernet of the S3c4510B). 
 
The observation presented here is with respect to larger files >= 1MB.
With smaller files (100k or less), the throughput in both
the directions (GET and PUT) were roughly the same.
 

What we tried:
--------------------
 
* We had increased the buffers of the Samsung ethernet driver
   for both the TX and RX but no noticeable improvement was 
   observed.
 
* We wrote an application on the embedded system (gateway)
  that simulated FTP. The client was on H1 and this application
  residing on the embedded system to receive the packets from 
  the client (on H1). The client sent packets at a high rate. 
 (Observation: Not a single packet was "overrun"). 
 
Our observations:
-------------------------
 
We had done several investigations and finally came to 
the following conclusions:
 
* The S3C4510B manual states that overrun can occur if 
  the buffers for receiving the packet are insufficient. This
  was increased sufficiently even to 128.
  Still the problem persisted.
 
* This problem could also occur on account of high
   system latency. We investigated this also and
   observed that there is no latency in the system.
 
* We finally concluded that there could be clash in
  the BDMA RX and BDMA TX of the S3C4510B ethernet 
  interrupt. Both share the same DMA and if there is a simultaneous
  occurence of both the events, then the prioritisation of these 
  could be an issue at higher data rates.
 
Looking forward to some help. Has anyone faced this scenario
before? any suggestions / solution to the problem?
 
Thanks in anticipation.
 
Thanks and with regards,
- Nagi
 

 


---------------------------------
Do you Yahoo!?
Free online calendar with sync to Outlook(TM).
[Attachment #5 (text/html)]

<DIV>Hi,</DIV>
<DIV>&nbsp;</DIV>
<DIV>I am sorry, my previous mail got inadvertantly posted before I completed \
it.</DIV> <DIV>&nbsp;</DIV>
<DIV>
<DIV>We've put uclinux onto an embedded system that is based on the</DIV>
<DIV>Samsung S3c4510B processor. This system has two ethernet</DIV>
<DIV>ports - one is provided by the processor (used for LAN port) and the </DIV>
<DIV>other is using a ethernet chip (&nbsp;AX88796 - ASIX used for WAN port). </DIV>
<DIV>The LAN chip is connected to the "external interrupt" line of the \
processor.</DIV> <DIV>&nbsp;</DIV>
<DIV>Everything as far as the system is concerned is ok. However, using this</DIV>
<DIV>box as a gateway between two networks (n1 and n2) we found</DIV>
<DIV>that there were some throughput problems. From a host H1(on N1)</DIV>
<DIV>we did a FTP&nbsp;put of a file to host h2(on N2). The throughput observed</DIV>
<DIV>was 30 KB/s. Instead of a FTP put we did a get and observed</DIV>
<DIV>a throughput of 550 KB/s. We were trying to debug this asymmetric</DIV>
<DIV>throughput problem. </DIV>
<DIV>&nbsp;</DIV>
<DIV>We observed that when doing a FTP put, we observed the output</DIV>
<DIV>of the "ifconfig" after a put and found around 15% of total packets</DIV>
<DIV>are "overrun" on the LAN port of the gateway (this is the internal</DIV>
<DIV>ethernet of the S3c4510B). </DIV>
<DIV>&nbsp;</DIV>
<DIV>
<DIV>The observation presented here is with respect to larger files &gt;= 1MB.</DIV>
<DIV>With smaller files (100k or less), the throughput in both</DIV>
<DIV>the directions (GET and PUT) were roughly the same.</DIV>
<DIV>&nbsp;</DIV></DIV>
<DIV>What we tried:</DIV>
<DIV>--------------------</DIV>
<DIV>&nbsp;</DIV>
<DIV>* We had increased the buffers of the Samsung ethernet driver</DIV>
<DIV>&nbsp;&nbsp; for both the TX and RX but no noticeable improvement was </DIV>
<DIV>&nbsp;&nbsp; observed.</DIV>
<DIV>&nbsp;</DIV>
<DIV>* We wrote an application on the embedded system (gateway)</DIV>
<DIV>&nbsp; that simulated FTP. The client was on H1 and this application</DIV>
<DIV>&nbsp; residing on the&nbsp;embedded system to receive the packets from </DIV>
<DIV>&nbsp; the client (on H1). The client sent packets at a high rate. </DIV>
<DIV>&nbsp;(Observation: Not a single&nbsp;packet was "overrun"). </DIV>
<DIV>&nbsp;</DIV>
<DIV>Our observations:</DIV>
<DIV>-------------------------</DIV>
<DIV>&nbsp;</DIV>
<DIV>We had done several investigations and finally came to </DIV>
<DIV>the following conclusions:</DIV>
<DIV>&nbsp;</DIV>
<DIV>* The S3C4510B manual states that overrun can occur if </DIV>
<DIV>&nbsp; the buffers for receiving the packet are insufficient. This</DIV>
<DIV>&nbsp; was increased sufficiently even to 128.</DIV>
<DIV>&nbsp; Still the problem persisted.</DIV>
<DIV>&nbsp;</DIV>
<DIV>* This problem could also occur on account of high</DIV>
<DIV>&nbsp;&nbsp; system latency. We investigated this also and</DIV>
<DIV>&nbsp;&nbsp; observed that there is no latency in the system.</DIV>
<DIV>&nbsp;</DIV>
<DIV>* We finally concluded that there could be clash in</DIV>
<DIV>&nbsp; the BDMA RX and BDMA TX of the S3C4510B ethernet </DIV>
<DIV>&nbsp; interrupt. Both share the same DMA and if there is a simultaneous</DIV>
<DIV>&nbsp; occurence of both the events, then the prioritisation of these </DIV>
<DIV>&nbsp;&nbsp;could be an issue at higher data rates.</DIV>
<DIV>&nbsp;</DIV>
<DIV>Looking forward to some help. Has anyone faced this scenario</DIV>
<DIV>before? any suggestions / solution to the problem?</DIV>
<DIV>&nbsp;</DIV>
<DIV>Thanks in anticipation.</DIV>
<DIV>&nbsp;</DIV>
<DIV>Thanks and with regards,</DIV>
<DIV>- Nagi</DIV>
<DIV>&nbsp;</DIV></DIV>
<DIV>&nbsp;</DIV><p><hr SIZE=1>
Do you Yahoo!?<br>
Free <a href="http://us.rd.yahoo.com/mail_us/tag/*http://calendar.yahoo.com">online \
calendar</a> with sync to Outlook(TM).



_______________________________________________
uClinux-dev mailing list
uClinux-dev@uclinux.org
http://mailman.uclinux.org/mailman/listinfo/uclinux-dev
This message was resent by uclinux-dev@uclinux.org

[prev in list] [next in list] [prev in thread] [next in thread] 

Configure | About | News | Add a list | Sponsored by KoreLogic