Re: [lwip-users] Improving LWIP performance

[Karthik Vadambacheri Manian]

Hi Kieran,

I investigated about the previous issue about interrupts and found the
fix in the driver code. The problem is RTL8139 uses level triggered
interrupts and the driver was not properly unmasking the interrupts in
a timely fashion so that new interrupts may be delivered.

Now I am trying to compare the performance of LWIP TCP/IP stack in
kitten LWK with TCP/IP stack in Linux both running in VM.  I have few
questions:

1) Can I avoid checksum generation & checkings as everything is
running in VMs? whats your opinion?
2) Can I compare the timings obtained for a pingpong using LWIP
sockets with normal linux sockets or should I compare against
nonblocking linux sockets. I am asking this because LWIP has tcpip
thread which polls for data so should I be using nonblocking linux
sockets and poll them to be sure I am doing a fair comparison. Please
let me know.

Also thanks to LWIP_STATS and Tuning TCP page in LWIP wiki. They
really helped me to improve the performance.

Thanks,
karthik