stress tests
[1/3] from: m::koopmans2::chello::nl at: 9-Oct-2001 21:20
Some preliminary results, for those who need to backup Rebol in corporate
environments.
Rebol (especially in fastCGImode on *nix) beats the competition ...
I figured this may be nice to know for all of you. We currently have Rebol /
Rugby tested by an external party.
Part of the routine in a major project with new technology.
The first results are promising: an echo function used with Rugby on a
dual 1.4 Ghz Athlon with Linux went to 2.5 million reqs in one and a half
hour. Roughly 450 / sec. Thumbs up for the non-blocking IO. Keep in mind
that
Rugby compresses/decompresses all messages and computes a checksum/secure
for message integrity, so it is quite a bit of logic itself.
We also did a denial of service attack scenario: the OS (Solaris and
Linux) is the bottleneck there. Once that stabilizes Rebol nicely returns.
People are... impressed.
--maarten
[2/3] from: petr:krenzelok:trz:cz at: 9-Oct-2001 22:03
Heh, that's cool to know - few questions here:
- what is your set-up? I mean - Apache, fast cgi, static, dynamic, external?
- what script is launched by Apache? client side of rugby, connecting to
already running rugby server/service?
- would it have any speed impact (increase), if you wouldn't use
checksummed, secured connection? As in my opinion, if external customer
reaches your site in https (ssl) mode, you don't imo need encrypted
communication on internal lines, or am I wrong?
Thanks a lot,
-pekr-
[3/3] from: gchiu:compkarori at: 10-Oct-2001 12:11
On Tue, 9 Oct 2001 21:20:12 +0200
"Maarten Koopmans" <[m--koopmans2--chello--nl]> wrote:
> The first results are promising: an echo function used
> with Rugby on a
> dual 1.4 Ghz Athlon with Linux went to 2.5 million reqs
> in one and a half
> hour. Roughly 450 / sec. Thumbs up for the non-blocking
> IO. Keep in mind
BTW, I don't think Maarten is suggesting that you stress my
chat server with 2.5M reqs :)
It so far seems to have done nicely with 4 people on
simultaneously. More would be nice for testing.
For each client sitting "connected" to the server, they send
one request every three seconds (polling). If they start
chatting, then each message is just another request.
--
Graham Chiu