tomcat-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jean-Luc Rochat <>
Subject Re: Discussion: AJP next
Date Wed, 02 Feb 2000 12:41:57 GMT wrote:
> >
> > 1 - integrating a JVM in an Apache module is a _bad thing_ (tm) IMO.
> >     . from the architecture point of view. (n-tier is better for
> > scalability than monolithic servers)
> >     . from the security point of view (small pieces of code means less
> > bugs, user permissions problems)
> >     . from the portability point of view (example: Apache runs very well
> > on BSDs. JVMs don't).
> >
> You are correct, integrating a JVM into apache (or any other web server) is not
> good. However, on NT there are performance improvements when working in-process
> and if you want to create the biggest bang this is the way to go.

I disagree here. If you want the fastest helloworld servlet, you can go
for it.
If you want something that could be used on a large scale of users, you
As soon as you have a useful servlet, the performance bottleneck under
load is not the http server, not the network (speaking of simple
requests), but the JVM. Putting it on a dedicated HW, or better
distribute the load between n JVMs * m hosts is the way to go. unless
you use a 3090 ;-)
But I here understand that it is not a goal for a servlet reference
implementation to support thousands of users.
I'm not against using an embedded JVM, if it's pluggable. I'm just
saying that scalability is still one of the reasons for people to choose
products to run their software.
> I agree that a JNI only solution is bad (does not scale/not as stable ...) but
> we should
> try to make it possible.
try to build non-portable java code ;-))  but OK, if it's a *possible*
choice, I'm all for it.
> >
> > 4 - sockets cannot be easily shared between httpd processes (on Unix at
> > least, and with Apache 1.x).
> >
> Neither on NT.
Can't share a fd on NT between threads in the same process ? hmmm...
inter-thread communications mechanisms should offer the creation of

> >
> > 5 - Using any other protocol than ajpv11 is faster.
> >     . yes but ajpv11 brings everything needed. So keep the "verbose"
> > protocol a possible choice.(for ben-ssl/mod_ssl by example).
> >
> We are not saying that we want to kill ajp12, only that we want to add something
> new.
I was not saying I wanted to use ajpv12. In fact I was trying to say
that I wanted to make this (pluggable protocols) possible.
> >
> > 7 - All of this has to pass firewalls (callbacks ?)
> >
> On the same socket.

>      Gal Shachor

View raw message