GlobalNOW Benchmark, initial results

Schilling, Richard RSchilling at affiliatedhealth.org
Fri Oct 27 15:59:02 EDT 2000


Well, it's been about a week since I announced the start of the GlobalNOW
benchmarking project, and I'm pretty sure some of you checked it out.  Thank
you all for taking a look at the web pages where the benchmarking is being
done (http://www.nationalinformatics.com and
http://www.affiliatedhealth.org).  The GlobalNOW benchmarking project is
designed to determine a simple, but useful metric of the potential
processing power that can be harnessed from the typical Java (tm) Virtual
Machine running on multiple browsers at the same time.  If you would like
more information about the project, check out the project's home web site at
http://www.nationalinformatics.com/GlobalNOW.


GENERAL OBSERVATIONS

The raw data can be found at
http://www.nationalinformatics.com/download/tblBenchmarkLog.txt in ASCII
delimited format.  After such a short time, it's difficult to conclude
anything earthshaking, but looking at the data at this point indicates a few
things.

* For the benchmarking to work at the National Informatics web site, it is
necessary for the browser's user to set permissions of the browser so that
the applet has access to all network addresses.  If not, a security
exception is generated.  

* The record breaker for calculations per second is without a doubt
connecting from www.appliedthermalsciences.com (IP 207.5.141.74) at a peak
rate of 189878 calculations per second.  Did the browser's virtual machine
get an injection of steroids?

* Beyond the appliedthermalsciences.com browser, the highest rate of
computation came from the local host (client address 127.0.0.1).  No
surprise here.  Be aware that the data includes some testing data, which can
be identified by the host address and remote address of the client (they're
both 127.0.0.1).  

* The data is comprised of 1,255 visits were recorded between 10/13 and
10/27.  The web site this was done on (http://www.affiliatedhealth.org) is
certainly not a high traffic web site.  So if this were done, say on the
NASA web site, we would expect better calculation rates.

* The only browsers that the project has captured are IE, and Java (which
may or may not be an actual web browser).  Netscape nor any of the other
brosers ran the benchmark for whatever reason.

* Some visitors to the web site have noticed slower performance.  This could
be due to higher web site traffic in general, but also due to the time it
takes the Java applet to load.  We are in the process of getting the
bandwidth upgraded, which will help me determine why the site is slower now.


RAW DATA

Here's a legend for the columns in the raw data file
(http://www.nationalinformatics.com/download/tblBenchmarkLog.txt):
ID				Unique identifier for the record
Column Name			Description
DocumentLocation	 	The web page that recorded this benchmark.
CGI_HTTP_HOST 		The address of the host.
CGI_REMOTE_ADDR		The client address.
CGI_HTTP_USER_AGENT	Name of the client (e.g Mozilla, IE4, Netscape).
BENCHMARK_NAME		Name of the benchmark taken
RUN_DATE			Date and time the benchmark was recorded
CALCULATIONS		Number of calculations performed.  NOTE: One
calculation is equivilant to about 15 operations (memory stores, additions,
memory references, etc. . .) in the algorithm.
RUNNING_TIME		The number of milliseconds the applet run
CALCULATIONRATE		The number of calculations per second
AVERAGE_MEMORY		The average amount of free memory in kilobytes
available while the applet was running
DEBUG				Indicates if DEBUG is set to ON or OFF.

The file is ASCII, character delimited (the delimiter is the vertical bar
"|").


THE GLOBALNOW PROJECT

GlobalNOW is the name I've given to the software technology I've developed
that allows an unlimited number of Internet browsing workstations to become
part of a Network Of Workstations (NOW) hardware cluster.  This is
consistent with the definition of a Beowulf class machine, which has a head
node and several child nodes running processes.  Under the GlobalNOW
strategy, the head node is the web server itself, and the child nodes are
ordinary web browsers running a Java (tm) Virtual Machine that visit the web
site.  The visiting browsers become nodes in the GlobalNOW cluster by
uploading a standard Java Applet from the web server, performing a set of
calculations and then returning the results to the web server via the HTTP
protocol.

It is theoretically possible to harness massive amounts of computing power
by using GlobalNOW technology on high traffic web servers.  The GlobalNOW
benchmark is designed to give everyone an idea of what kind of power to
expect were we to roll out the technology on a high traffic web site, such
as the NASA web site.  Using Java applets in this way is a safe method of
attaining processing power instantly.  The security model used by Java
applets makes it possible to perform reliable, secure processing using any
web browser that correctly implements the Java Virtual Machine.  Also, there
is no hardware to install (unless you need a new web server), and virtually
no cost to develop the software, save people time.  



Let me know if you have any questions.

Richard Schilling
Webmaster/Web Integration Programmer
Affiliated Health Services
Mount Vernon, WA
rschilling at affiliatedhealth.org
http://www.affiliatedhealth.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.clustermonkey.net/pipermail/beowulf/attachments/20001027/e12da734/attachment.html>


More information about the Beowulf mailing list