Wow - excellent response! Thanks much for the additional benchmark
information. It looks like you guys have done a very impressive job on
Orion. Congratulations.
We're considering using EJB's (along with JSP's and servlets) so
OrionServer might serve (no pun intended) our purposes there, as well. High
scalability, robustness, and session-replication (i.e. if one server gets
taken down the other servers should be able to take over with all of the
necessary session information) are important to us. I don't know how
OrionServer compares in those areas but I'll take a look.
Thanks again,
Brien Voorhees
----- Original Message -----
From: Karl Avedal <[EMAIL PROTECTED]>
To: Brien Voorhees <[EMAIL PROTECTED]>
Cc: <[EMAIL PROTECTED]>
Sent: Tuesday, October 12, 1999 3:22 PM
Subject: Re: JSP customized tags (also ORIONSERVER, JRUN, BENCHMARKS)
> Hello Brien,
>
> > OrionServer (www.orionserver.com) also supports JSP1.1. I've been using
> > JRun which supports custom tags but I believe it's using their own
> > proprietary method. Does anyone know if JRun is going to support the
> > customized tag standard anytime soon? I'm considering switching from
JRun
> > to OrionServer.
>
> I don't know when JRun will suport it, but thank you for considering our
product.
>
> >
> > On a side note, the orionserver site has some impressive (almost
> > unbelievable) benchmark
> > comparisons(http://www.orionserver.com/benchmarks/benchmark.html)
against
> > ServletExec. Has anyone done comparisons(especially against JRun)?
>
> Yes, I did some testing today (I am however with the Orion team, so I'm
very
> partial, but anyway)
>
> Today's testing session was set up like this:
>
> I installed Orion 0.7.6b, Weblogic 4.5.1, Resin 1.0 and JRun 2.3.3 on the
same
> machine (a PII-266 Mhz machine with 96 Mb RAM). I downloaded the benchmark
files
> (http://www.caucho.com/articles/benchmark.jar) Caucho have used for
testing Resin
> with (in an attempt to get more fair results).
>
> Also, to get rid of the problem of people thinking the web server is the
> bottleneck, I didn't install any of the engines as plugins to another
webserver but
> used them directly.
>
> I used the latest JDK (1.3b1) with the latest Hotspot (2.0 early access)
(getting
> some servers configured to use this was a hassle)
>
> To start with I tested the hello_java.jsp file in the benchmark package.
>
> This was my testing cycle for every engine:
>
> 1. Shut down all running applications
> 2. Start the server
> 3. Let httpbench (downloadable with source from
http://www.orionserver.com) hit the
> hello_java.jsp for 2 minutes to get hotspot going.
> 4. Start httpbench again, let it run for 1 minute. Take notes of the
result of
> pages/sec and latency
>
> These four steps was repeated three times for every engine and an average
was
> calculated.
>
> These were the results:
>
> JRun 2.3.3: pages/sec: 25.68579867 latency: 192.7007383
> Weblogic 4.5.1: pages/sec: 134.1389787 latency: 37.172157
> Resin 1.0: pages/sec: 209.751607 latency: 23.64778167
> Orion 0.7.6b: pages/sec: 246.820906 latency: 19.270431
>
> Now you might of course ask why the difference is this huge. I don't know
much
> about the other engines, but I do know that our developers have worked
very hard to
> make Orion perform well, and I guess that is what is paying off.
>
> > I
> > suspect that they're achieving such a large benchmark margin by
optimizing
> > out unnecessary jsp varable setup which might work for simple pages but
not
> > for any real pages. By unnecessary jsp variables, I mean code like the
> > following in the beginning of a jsp-generated java file :
> >
> > ServletConfig config = getServletConfig();
> > ServletContext application = config.getServletContext();
> >
> > Object page = (Object) this;
> > PageContext pageContext =
> > JspFactory.getDefaultFactory().getPageContext(this, request, response,
> > null, true,8192, true);
> > JspWriter out = pageContext.getOut();
> >
> > HttpSession session = request.getSession(true);
> >
> > For a very simple (i.e. unrealistically simple) jsp that doesn't
reference
> > session or request variables that code could be omitted. This is pure
> > speculation on my part - I have no idea if that's what they're doing but
> > it's the only thing I can think of to explain their benchmark claims.
> > Anyone have any real comparison info?
> >
>
> To try this I also ran tests on session_java.jsp (which sets and gets a
session
> variable) to see if this could have anything to do with it. I only did
this test
> for Orion 0.7.6b and the result was 222.414375 pages/sec. That is slower
than
> without using a session but still faster than any of the others managed
without a
> session.
>
> >
> > I just took another look at the benchmarks page and noted that they were
> > using Microsoft Personal Web Server as the server to plug ServletExec
into.
> > Could it be that MPWS is really the bottleneck?
> >
>
> As I mentioned before, I made this test without any web servers to get rid
of any
> problems there might be with web servers destroying the benchmark.
>
> I'd love to see other people do benchmarks of this. It feels great to do a
test and
> see that your product beats the others, but it would feel so much greater
if the
> test was done by someone else :)
>
> Usual benchmark disclaimer: Remember, a benchmark is always a benchmark
and you
> should always be sceptic. What you should do is deploy your own JSP:s in
different
> servers and see what performs best for you. That is the only way to know
for sure
> what is best for you.
>
> Karl Avedal
> The Orion team
>
>
===========================================================================
To unsubscribe: mailto [EMAIL PROTECTED] with body: "signoff JSP-INTEREST".
FAQs on JSP can be found at:
http://java.sun.com/products/jsp/faq.html
http://www.esperanto.org.nz/jsp/jspfaq.html
