Re: Java Benchmarks - AS/400 Kicks - A#$

[dhandy@xxxxxxxxxxx (Douglas Handy)]

Leif,

>Clearly someone got some numbers screwed up here casting doubt on the whole
>thing. Also, the first benchmark used 200 connections, the next 9000
>connections.
>Something is not right here. 

I think the whole point was the scalability issue, and how the AS/400
stood up better as the number of connections increased.

Note the chart at the end with various timings for the 12-way as the
number of network connections increased.  In order to sustain the
20,000 connections the JVM had to handle 40K threads concurrently
which is allegedly more than Sun has ever accomplished.

I think it would have been helpful to print Sun's results as well at
the various connection counts.  The presumption here is that Sun's
machines performance knee is between 200 connections and 9000, and
that the overhead of trying to handle 9000 connections (18K threads?)
causes the Sun to spend to much time in task swapping and dispatching
vs actual run time.

At only 200 connections, the per thread overhead disparity would make
much less difference.  So the numbers may not be quite as suspect as
they may appear on the surface.  It sure would have been helpful to
see more of Sun's actual figures though.

Maybe there is a real-world need for 40K concurrent threads, but it
sure isn't at any of the clients I have ever worked at ...

Doug
+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---