-----------------------------
Please read the FAQ!
<http://java.apache.org/faq/>
-----------------------------
> So in the total throughput test, the difference narrows (JServ is only
> 33% slower than JWS as opposed to 75% slower in the first test).
>
> My simple question is why?
>
> The only thing I can think of is that there is some sort of TCP/IP
> bottleneck. In TestServlet2 I am accessing the DB to get the bytes (via
> TCP/IP). When using JServ because of its architecture of being a
> separate process from Apache also using TCP/IP, I need to move the bytes
> a second time (really three times if you count the response back to the
> client), once from Oracle to JServ, a second time from JServ to Apache.
> In JWS the bytes only are only moved once (from Oracle to JWS).
> Therefore if there is some bottleneck in the TCP/IP loopback (since this
> is all on the same machine, the bytes aren't really going out to the
> network, except of course when they get sent back to the client), the
> JServ architecture might be inherently slower than JWS for servlets
> doing database access to a local database.
>
> thanks,
> --Barry
You might be onto something Barry. I read some recent articles about the
performance shoot out
between Linux/Apache and NT/IIS. They redid the test using Red Hat support
techs for Linux tuning.
It was discovered during the test that the Linux kernel is currently using a
single-threaded IP stack.
It was concluded that this was the reason the NT/IIS system was faster than
the Linux/Apache one.
This might also be the reason for your performance difference, using more IP
calls. There is
supposed to be an effort to create a multi-threaded IP stack, but I can't
attest to it.
Ed.
--
--------------------------------------------------------------
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]
