Sample NetLogger Results


Here you can find a collection of previous NetLogger results, including log files, nlv-keys files, and sample screen dumps. If you are trying to figure out how to display your data in NLV, looking at these examples should give you some ideas.

NetLogger Example
Description
screen dumps
raw data and nlv_keys files
Amber and Ganglia
Data From Amber application (SC2003 demo). Includes CPU data.

nlv results

download

Remote IO application (DPSS)

This shows the results for a NetLogger Analysis of High Energy Physics analysis application (STAF) doing a remote read of data on a DPSS server. The blue, green, and brown lines are tracing a data read request as it is send to the remote server. The lines fan out as a single read request on the client get split into multiple disk reads on the server. In this case data resided on 4 DPSS servers, hence the 4 colors of lines. The red lines show what is happening inside of the STAF application. "PAM" is a CPU intensive operation, shown at the top of the screen.

What we learn from this plot: This application spends about twice as much time doing data analysis as it spends doing remote IO. Overlapping I/O and computation would result in a large increase in performance.

nlv of STAF/DPSS 


nlv of STAF/DPSS with analysis window
: This slows a zoomed in view of the above plot, along with the nlv analysis window

nlv of DPSS client with many TCP streams

download
FTP application (Globus GridFTP)
This is an example of using NetLogger to analyze and debug the Globus GridFTP program. These results are client side only.
Note the large overhead (about 9 seconds) of the initial FTP handshaking.

nlv of Globus IO  

nlv of Globus showing unbalanced streams bug

download
Visualization of Remote Data
This is an example of how NetLogger was used to tune a visualization application (Visapult) reading remote data from a DPSS. Before adding NetLogger, the application sequentially did IO, followed by computation. After seeing these results, it became clear that overlapping IO with computation would help a lot.

nlv of Visapult/DPSS; before tuning : (sequential IO and computation)

nlv of Visapult/DPSS after tuning : (overlapped IO and computation)


Network Monitoring (NTAF)
Example of using NetLogger to correlate the results of several network tests
nlv comparing iperf to pipechar results

download
web100 Monitoring (iperf)
Example of how NetLogger can be used to analyze the results of iperf running on a web100 Linux kernel
iperf + web100

more iperf + web100
download

More raw data and examples may be available here.