[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [lwip-users] lwIP 1.1.0 released
From: |
Leon Woestenberg |
Subject: |
Re: [lwip-users] lwIP 1.1.0 released |
Date: |
Sun, 02 Jan 2005 19:14:25 +0100 |
User-agent: |
Mozilla Thunderbird 0.7.1 (Windows/20040626) |
Hello Jim,
Jim Gibbons wrote:
Could we prevail upon you for a "state of lwIP" note? It could be
seen as timely at the New Year, and with this new release. It's hard
for me, and perhaps for others, to accurately judge the big picture
from the daily flow of detail oriented messages.
Good point.
Could you tell us what you think is truly stable and good? where you
think the development frontiers lie? and perhaps where lwIP should
not be taken?
From my news announcement on Savannah:
"Release 1.1.0 improves TCP throughput performance, fixes some small
bugs in TCP, and annoying bugs in the ARP cache and ARP packet queueing.
Check the CHANGELOG for details."
In general, this is my impression:
The lwIP 1.1.0 core is as stable as 0.7.2 in general, which has proven
production quality.
It is more correct/stable regarding TCP operation, as well as more
performant in TCP.
Outgoing packet queueing is stable (but limited to one packet per ARP entry,
the multipacket implementation clashed with TCP's queue.)
The core of lwIP uses the raw (callback) API and udp_(), tcp_()
functions. In a multitasking
environment, treat each lwIP call as a critical section, i.e. *serialize
access to the lwIP stack*.
This will be safe.
The socket layer (blocking calls and the sysarch layer) - I really
cannot tell. I have never used nor
maintained it. (I do not fancy its design, buts that's a personal opinion).
Regarding development frontiers, (outgoing) packet queueing is quite new
and I may look again at
the problems surrounding multiple packets. TCP performance optimization
has been tickled a few
times, we may revisit this once again. Rendez-vous is something we might
do, it might be nice for
embedded devices.
Then again, m personal opinion is that lwIP should not grow much beyond
what it is now. BSD
stacks are right around the corner with the 32-bit MMU processors
becoming increasingly
interesting/cheap for embedded applications.
Personally, I am interested to see what lwIP does in eCos (Jani Monoses
has done much work
there).
Regards,
Leon Woestenberg.