From 0ea5fc66924303d1bf73ba283a383e2aadee02f2 Mon Sep 17 00:00:00 2001 From: neodarz Date: Sat, 11 Aug 2018 20:21:34 +0200 Subject: Initial commit --- pipermail/nel/2001-July/000488.html | 97 +++++++++++++++++++++++++++++++++++++ 1 file changed, 97 insertions(+) create mode 100644 pipermail/nel/2001-July/000488.html (limited to 'pipermail/nel/2001-July/000488.html') diff --git a/pipermail/nel/2001-July/000488.html b/pipermail/nel/2001-July/000488.html new file mode 100644 index 00000000..e59240d6 --- /dev/null +++ b/pipermail/nel/2001-July/000488.html @@ -0,0 +1,97 @@ + + + + [Nel] TCP vs. UDP + + + + + + +

[Nel] TCP vs. UDP

+ David Mentre + David.Mentre@inria.fr
+ 06 Jul 2001 09:54:59 +0200 +

+
+ +
Hi,
+
+>From all other posters, it seems that performance result is highly
+dependent on implementation. So to make the Really Good Choice(tm), you
+should have both and choose according to *figures*. So I would do that
+in the following order :
+ 
+ 1. design a "packet sending" layer which allows messages
+    sending/receiving with semantics in mind (see Vincent's post in the
+    same thread). Maybe NeL has already that, I haven't check the
+    source, sorry.
+
+ 2. implement this layer using TCP. Should be straightforward.
+
+   => gives you a backup implementation
+
+   => gives you basic performance figures
+
+ 3. design and implement UDP implementation_s_ with various
+    requirements depending on the following parameters:
+
+     - packet loss or not
+
+     - in order delivery or not
+
+    You could start from the easiest (packet loss && out of order <=>
+    raw UDP) and incrementally develop more complicated ones.
+
+ 4. substitute parts of TCP implementation (chat, position packets, ...)
+    with the corresponding UDP based protocol
+
+ 5. implement a benchmark and test, test, test, ... ; then go to step 4
+    and loop
+
+
+Ok. I don't propose any line of code so you could directly throw this
+post to /dev/null but you asked for comments. ;)
+
+BTW, what is your Linux configuration? I've been stuck with the need to
+reinstall everything (X, STLport, freetype) on my debian machine. So it
+may be easier to install the Linux distribution your are using at Nevrax
+on some spare partition to try to "enter the loop".
+
+Best regards,
+d.
+-- 
+ David.Mentre@inria.fr
+ Opinions expressed here are only mine.
+
+
+ + + + + + + +
+

+ -- cgit v1.2.1