Hi,
I can confirm that the latest refresh on IBM's JVM on Linux has networking
problems fixed. And its speed is good, the JSDK2.0 servlet engine serves
SnoopServlet at least three times faster than the Blackdown JDK1.1.7_v3,
green_threads.
(well, it was a stripped down SnoopServlet, taking out some of the code -
I wanted to have an image on servlet engine's speed not on
the servlet itself... e.g. I took out the getRemoteHost() wich involves
name resolving, getPathTranslated(), getInitParameterNames/values() -
their place is in init() and should not affect service() performance)
I ran tests with apache's benchmark "ab", wich I found in Apache 1.3.6
distributuion. Tests involved betwen 20 and 50 threads requesting
simultaneously the same page. It prooved to be very "agressive",
requesting far less CPU clocks than the Apache itself server when I
benchmarked Apache serving static pages.
The hardware I had were two boxes with K6II/350 Mhz, 128MB/100Mhz RAM,
Realtek 8139 based ethernet cards (used at 10 MHz, half duplex)
Operating system was Slackware 4.0; The big problem with it was to compile
the glibc 2.1.1 libraries, wich IBM's JVM requires.
One was the "clients", the other the server.
Some of the figures:
- Apache httpd, with some optimisations indicateded in its docs was
able to serve small pages (1-4k) at 1000-1200 pages per second.
- Apache's included cgi-bin/test-cgi, a shell script wich does
pretty much what SnoopServlet does - was served at around
70-75 responses per second.
The following were obtained with JSDK2.0 servlet engine - wich
I think it is fast enough, a faster (single-CPU) servlet engine
cant be too much faster.
- Blackdown's JVM 1.1.7 v3, green_threads ( could not make
the native_threads Blackdown JVM run on my system :-( ),
was pretty slow, 75-85 responses per second.
Not very big difference from interpreted /bin/sh CGI
- IBM's JVM, jdk 1.1.6 served 250 respones per second, a pleasant
surprise.
- The older JVM from IBM, the one with problems on networking served pages
at a rate of 320-400 per second, unbelivable fast but... the benchmark
utility reported that 25%-40% of requests were errors.
I have not ran pure JVM tests (mathematical operations, object
instantiation, String manipulation, all such stuff) nor JDBC tests,
with IBM's engine to give more hope for Linux fans that it will be pretty
soon a very performant Java environment.
Regards,
Cezar Totth
On Wed, 1 Sep 1999, Craig R. McClanahan wrote:
> Chris Pratt wrote:
>
> > I would like to try it, but I've heard that it has bugs related to running
> > servlets. If that's not true, I'll have to give it a try. What servlet
> > engine are you running?
> > (*Chris*)
> >
>
> I'm using the "refreshed" (last week sometime) version of IBM's JVM from the
> AlphaWorks site. I've used it lightly with Apache Jserv, and quite heavily
> using the JSWDK 1.0 final release.
>
> The earlier release of the JVM did indeed have some socket-related bugs that
> caused problems with moderate numbers of simultaneous hits. The refreshed
> release seems to have fixed those.
>
> Craig McClanahan
>
> ___________________________________________________________________________
> To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
> of the message "signoff SERVLET-INTEREST".
>
> Archives: http://archives.java.sun.com/archives/servlet-interest.html
> Resources: http://java.sun.com/products/servlet/external-resources.html
> LISTSERV Help: http://www.lsoft.com/manuals/user/user.html
>
___________________________________________________________________________
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff SERVLET-INTEREST".
Archives: http://archives.java.sun.com/archives/servlet-interest.html
Resources: http://java.sun.com/products/servlet/external-resources.html
LISTSERV Help: http://www.lsoft.com/manuals/user/user.html
