SCL Cluster Cookbook
Running NetPIPE

·   Getting NetPIPE   ·   NetPIPE Testing Strategies   ·   Graphing NetPIPE Results with gnuplot   ·   Graphing NetPIPE Results with Excel   ·   Interconnect Overview   ·   Interconnect Performance   ·
·   Cluster Cookbook Home   ·   Scalable Computing Lab   ·

Getting NetPIPE

To use NetPIPE to benchmark a network interconnect on UNIX systems, download the source file package netpipe.tar.gz to your system. Unpack and build the sources as described in the README file.

NetPIPE can be also be used on Windows NT systems. You can download the NT version of NetPIPE at ftp://ftp.scl.ameslab.gov/pub/netpipe/NT/ . The file nptcp.exe in that directory is a prepared executable file for testing TCP on NT, or you can download the source and it will (hopefully) build in the Visual C++ environment.

You will need two systems, a sender and a receiver, connected by the network interconnect that you wish to test. TCP/IP will have to be working between the systems over the interconnect. If necessary, copy the NPtcp executable program so that the NPtcp executable is present on both the sender and the receiver.

On the receiver system, execute the command

NPtcp -r

to start the NetPIPE receiver.

On the sender system, execute the command

NPtcp -t -h IP-address -P -o output-file

where IP-address is the IP address of the receiver and output-file is the name of the output file where you would like NetPIPE's data to be stored.

NetPIPE Testing Strategies

Here are some ideas for interconnect testing that can be done with NetPIPE:

Graphing NetPIPE Results with gnuplot

gnuplot is a freely-available graphing program that can be used on most UNIX systems. If it is available on your system, you can use it to plot NetPIPE data. To plot throughput versus block size, use the gnuplot command:

plot "netpipe-file" using 4:2

to plot the data from NetPIPE's data file netpipe-file. To plot throughput versus time (the network signature graph), use the gnuplot command:

plot "netpipe-file" using 1:2

It is useful to use the gnuplot command set logscale x on NetPIPE graphs, as the block size used for NetPIPE transfers increases on an exponential basis.

Graphing NetPIPE Results with Microsoft Excel

NetPIPE results can be imported into Microsoft Excel as columns of text and the resulting spreadsheet may be used in a chart. With Excel running, import NetPIPE data by selecting Open from the File menu and then change the Files of Type from Microsoft Excel Files to All Files (*.*). Then, use the directory navigation tools to locate your NetPIPE output file and use the Open button to open the file. At this point, Excel should recognize that the file is a text file and will present the Text Import Wizard.

At the first step of the Text Import Wizard, select the button Delimited, as the data is not in rigid column format. Click Next and at step 2, click the Space box in the Delimiters pane. Click Next, and the General data type should be OK for each of the columns, so click Finish. The NetPIPE data should then appear in your worksheet.

On the left side of the sheet, click the 1 in the box to select the whole first row. Select Rows from the Insert menu to insert a blank row at the top of the sheet. Enter the words Time, bps, Bits, Bytes, and Variance in cells A1 through F1 to title each column.

Now, create a graph of the data. Select Chart - On This Sheet from the Insert menu. At the point where you would like to place the graph in the spreadsheet, hold the mouse button down and move the mouse to make the rectangle the size of the graph that you would like. Then, the Chart Wizard window will appear.

In the Chart Wizard's first step, select the Time and Throughput columns (columns A and B) as the range to graph (don't include the Time and bps labels at the top of the columns in the range). In the next step, select XY (Scatter) as the plot type. In the third step, select 2 (lines with points) as the format. At the fourth step, make sure the Data Series is in Columns, the First 1 Column(s) are used for X Data, and the First 0 Row(s) are used for Legend Text. The sample graph shown will look kind of funny at this point with nearly all the points bunched up against the left side of the graph - this is OK and will be taken care of later. Go on to the fifth step and enter a title for this graph, then enter Time (Seconds) for the Axis Title: Category X and enter Throughput (Mbps) for the Axis Title: Category Y. Finally, click Finish and your graph will appear.

Double-click on the graph to edit it. Click on the X axis (the bottom edge of the graph), click the right mouse button to bring up the pop-up editing menu, and select Format Axis... from the menu. Click the Scale tab and click the Logarithmic Scale button. You will also have to adjust the Value (Y) Axis Crosses At field (to the smallest X value on the graph) using this pane to get the Y Axis to stay at the left edge of the graph. Click OK to finish formatting.

Now, the markers on the graphed line may be so close together that they obscure the data. To eliminate the markers, click on the graphed line, click the right mouse button to bring up the pop-up editing menu, and select Format Data Series... from the menu. Select the Patterns tab, click the None button in the Markers pane, and click OK.

To change the legend's label for the graphed data, click on the graphed line, click the right mouse button to bring up the pop-up editing menu, and select Format Data Series... from the menu. Select the Name and Values tab, enter a short descriptive legend (such as Fast Ethernet) in the Name field, and click OK.

Now, the Excel graph is basically done. Be sure to use the Save As item in the File to save this worksheet as a Microsoft Excel Workbook so you don't lose your work! You can go on to use the various formatting features and add things to the graph to make it look the way you would like. You may even want to insert another page in the worksheet and move the graph over to that page so that you can easily print the graph by itself.


Questions or comments? Send mail to ghelmer@scl.ameslab.gov
Copyright © 1997, 1998 All Rights Reserved by Scalable Computing Laboratory.
Disclaimer Notice
Maintained by ghelmer / Last updated on 06/10/98