----------------------------------------------------------------
BEFORE YOU POST, search the faq at <http://java.apache.org/faq/>
WHEN YOU POST, include all relevant version numbers, log files,
and configuration files. Don't make us guess your problem!!!
----------------------------------------------------------------
On Sat, 23 Oct 1999, Pierpaolo Fumagalli wrote:
> Jon Smirl wrote:
> >
> > I just noticed this report from Mindcraft:
> > http://www.mindcraft.com/whitepapers/iws/iwsee4-sh242.html
> >
> > Any comments on why iPlanet is beating Jserv?
> > Is iPlanet directly linked to the JVM, in other words, is the socket
> > connection from Apache to JServ the problem?
> >
> I don't know how iPlanet is structured, but, IMVnHO the problem resides,
> yes, in the socket connection. With the new version of JServ (1.1) we
> should be able to gain that margin.
If you look at the test closely, you will note they didn't compare
any servlet code more complex than something spitting out what
amounts to static content.
The static content, SHTML and CGI results had pretty minimal performance
differences. Sure, you can say that 15% performance difference is
significant. But the difference between being able to server 20 million
hits a day and 24 million hits per day is completely insignificant in the
real world.
The only test they did that actually used jserv (they give numerous
different reasons for not doing other tests) was the following test:
100% Java Servlet
We created this test to measure the performance of Java servlets.
All of the requests are made to the same Java servlet, which simply
returns 6250 bytes of HTML.
Hmm. I don't think I'll worry about how fast jserv is on that,
since I normally don't use servlets to serve static HTML. This is
the very sort of test where the overhead of jserv 1.0's socket
connections are quite overstated, sine the JVM isn't doing any
"real" work. On a more realistic system, the jvm would be doing
significantly more complex work (hey guys, that is the point of
sevlets and is where Java is such a win...), resulting in the socket
latencies being far less significant. You can see the impact of
something in the fact that the request latency starts at a low of
50ms for jserv vs. <10ms for iWS.
When the "Sun-Netscape Alliance" posts their bragging on their
site, they just happen to say "oh, we tested all these things like
JSP pages and iPlanet was always faster" without mentioning that a bunch
of the things they talk about weren't tested on any server other than
iPlanet.
And they happened to not be able to run any SSL tests on stronghold due to
a "handshaking bug" with their benchmark. Well gee, you would think that
if someone was doing a test they would spend a little bit of time to
actually get more than half the results.
They also, of course, wouldn't think of testing against a C module,
as opposed to a C CGI.
They also take great pains to put up the config files to show how fair
they are, but they neglect the little detail of the jserv config files.
I also note that they misspell "JServe". Hey guys, when you can't get
the name of the product you are testing right...
--
--------------------------------------------------------------
Please read the FAQ! <http://java.apache.org/faq/>
To subscribe: [EMAIL PROTECTED]
To unsubscribe: [EMAIL PROTECTED]
Archives and Other: <http://java.apache.org/main/mail.html>
Problems?: [EMAIL PROTECTED]
