Holger, anyone? TCP performance (was) Re: [testing rugby-core]
[1/2] from: petr::krenzelok::trz::cz at: 19-Feb-2002 9:11
Hi,
I don't want to talk Rugby here now, but let's look at figures below once again
please. I first thought the slow Rebol tcp communication concerns only Czech W95 +
W98, but Pat is surely not using Czech Windows.
pat665 wrote:
> Hi petr
>
> Following your request, I have tested rugby-core versus rugby-xpi. Here are
> the results :
>
> rugby-core (http) -> duration 0:00:59.98
> rugby-xpi (tcp) -> duration 0:00:41.85
> rugby-xpi (http) -> duration 0:00:40.7
>
So, loop of 100 echos lasts some 40 sec on W9x, while on W2K it is some 100 - 150
request pers second! I would be really interested what is happening with W9x TCP
stack, and why it lags. Does it use delayed ACK? Does it really matter what
technique it uses in the background? Maybe more ppl here could try below mentioned
Rugby tests?
console1:
do %rugby.r
serve/with [echo] tcp://:8005
console2:
do %rugby.r
do get-rugby-service http://localhost:8005
start: now/time/precise loop 100 [echo "test"] print now/time/precise - start
I can't believe the lag is caused by Rugby architecture, it has to be something
under the hood of Windows API, and maybe some set-modes could be used for tcp
ports to tweak the performance ....
Thanks a lot,
-pekr-
[2/2] from: koopmans:itr:ing:nl at: 19-Feb-2002 12:25
Who did the test on W2K then?
I do Linux, and I can't recall having seen W2K?
IIRC W2K has a brand new TCP/IP stack indeed. OTOH, I didn't have those
problems on WinME. Sigh.
--Maarten