[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Ethernet thoughput limited

Unfortunately, for our application we need to use 2216 as our send size.  We
are now using non-blocking sockets, but are still seeng "spikes" of up to
3.5 msecs per send.  Could there be another process preempting the send?  We
are setting the send thread to be the highest priority thread, and the
overall process is set to the highest priority.

Any further advice you have would be appreciated.

Aaron R. Polichar
Ph:     858-826-9990
Fax:    858-826-9718
e-mail: polichara@xxxxxxx.com

-----Original Message-----
From: Mikael Starvik [mailto:mikael.starvik@xxxxxxx.com]
Sent: Tuesday, September 10, 2002 7:06 AM
To: 'Polichar, Aaron R.'; dev-etrax
Cc: 'Marlin, Shawn R.'
Subject: RE: Ethernet thoughput limited


2216 bytes is a suboptimal transfer size when testing 
performance. 2216 bytes is rougly 1.5 Ethernet packets 
and will therefor generate more packets on the network
than necessary (unless the TCP/IP stack in the sender
rearranges the data). Try with a larger size in each
transfer (e.g. 64 KB).

-----Original Message-----
From: owner-dev-etrax@xxxxxxx.com]On">mailto:owner-dev-etrax@xxxxxxx.com]On
Behalf Of Polichar, Aaron R.
Sent: Friday, September 06, 2002 8:48 PM
To: dev-etrax
Cc: Marlin, Shawn R.
Subject: Ethernet thoughput limited


We are having a problem getting the expected network throughput from the dev
board.  We are using the ETRAX100LX dev board on a dedicated network,
running at 100Mbps, full duplex.  We are sending 5000 packets to a Windows
2000 PC.  Each packet has a length of 2216 bytes, but we are only seeing a
max throughput of around 16Mbps.  We would like to get closer to 25-30 Mbps.
On the Windows PC, we have set the priority of our reader to realtime, and
we have set the priority of the process on the dev board to the maximum
priority.  We ran the same test on our corporate network with a PC running
Linux in place of the dev board, and we saw a throughput of around 70 Mbps.
I have turned off the "Fast serial port DMA flush" option with make xconfig.

Aaron R. Polichar
Ph:     858-826-9990
Fax:    858-826-9718
e-mail: polichara@xxxxxxx.com