RE: [lwip-users] Performance results

[John Taylor]

>From: address@hidden
[mailto:lwip->address@hidden On
Behalf Of Kieran >Mansley
>
>... but it does demonstrate that the stack can run fast (if you 
>rip out all the slow bits!)

What are the "slow bits?"  

I am finishing up porting lwip to two Hitachi H8s eval boards (edosk2674
@33Mhz and esp-hbdb01 @20Mhz).  I noticed that lwIP had what appeared to
be a slow response time to pings (1.3 msec).  I thought at first it was
because of my debug environment, but I ran in "stand-alone" mode and the
times did not change.  

Then I thought, well maybe its how I implemented my low-level Ethernet
driver.  So I ran my port of Adams Dunkels' uIP on the same boards.  My
ports of uIP and lwIP have identical low-level drivers - with the only
exception being that lwIP uses PBUFs and uIP has a single global buffer.
Okay, there is more indirection with the PBUFs than with an absolute
address, but the ping times for uIP are .8msec.  The ping times for lwIP
are 65% SLOWER than uIP.  I find it hard to believe this large
difference is due to PBUFs vs. a static buffer.  Both boards have a
16bit MCU with 32bit general purpose registers and the same compiler and
same RTOS was used for all of the test apps. The test apps for both
stacks are web-servers with no network traffic except for the pings.

One last caveat.  My port lwIP is not finished.  I still have bug where
the TCP "ack" number gets out-of-sync and effectively stalls some of the
connections.  However, I can flood ping the board and not loose a single
packet.


Is a 1.3msec ping response time reasonable for lwIP or have I screwed
something up in my port?



_______________________________________________
lwip-users mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/lwip-users