Network Characterization Service (NCS)

Computational Research Division
Lawrence Berkeley National Laboratory



Tools based on hop-by-hop network analysis are increasingly critical to network troubleshooting on the rapidly growing Internet. The network characterization service (NCS) provides the ability to diagnose and troubleshoot networks hop-by-hop in an easy and timely fashion. It includes the tools netestand pipechar. Using NCS makes applications capable of fully using the high-speed networks, e.g., saturating 1 Gb/s local network from a single x86 platform.

The algorithms used in release 1.3 (not BETA or earlier) to measure available bandwidth have been mathematically proved to be accurate and non intrusive <2002-12-08> . See results from emulation network Currently, it is implemented for end-to-end measurement in netest revision 2. The hop-by-hop implementation will be in the formal NCS 1.3 release.

This available bandwidth algorithm privodes a new mechanism for designing new network transmission protocol Network Lion .


Documentation:


API:


Download links:   Available for download are both binaries and source code. If you have any problems, please contact DAAgarwal@lbl.gov.

Supported platforms:

NCS is currently tested on following platforms:
  • FreeBSD (best performances with on NIC timer support)
  • BSD/OS and possible all other BSD O.S.s
  • Linux
  • Mac OS X
  • Solaris
  • AIX
  • IRIX
future support:
  • Digital UNIX
 
It does not and will not run on T3E because the T3E compiler has no 16-bit integer type.

The generic NCS functions should be able to compile and run on platforms that comply with IPv4 standard. Only kernel timer related functions need the FreeBSD KLD mechanism, therefore kernel timer related functions are only available on FreeBSD platform (Mac OS X and other BSD OSs are possibly supportable).
However, the on NIC timer feature can be supported for Linux if the sysKonnect driver supports this function under Linux.



Emulation testbed results

Emulation Network Topology

Emulation testbed results

 

Utilization % (loss %)

run time (sec.)

netest results

Accuracy

 

Pathload

GigE network

MTU = 9K

 

50~100 tests per run (300 sec.)

(require longer measurement duration)

available bandwidth (Mb/s)

(%)

 

Mb/s

11-13 seconds

<possible means guessing>

0 (0)

 

 

 

2.4 - 6.5

 

including MBS measurement

 

maximum throughput: 851

 

N/A

10 (0)

possible > 823

20 (0)

791.0 - 791.2

98.875

possible > 782

30 (0)

690.0- 691.0

98.643

possible > 678

40 (0)

598.5 - 599.0

99.750

possible > 567

50 (0)

502.5 - 502.9

99.420

possible > 480

60 (0)

403.8 - 403.9

99.025

possible > 350

70 (0)

306.4 - 306.6

97.833

possible > 262

80 (0.01)

7.89

(11.9)

210.0 - 211.0

205

94.500

97.500

possible > 217

90 (0.01)

13-15

(26)

113.0 - 115.0

102

86.000

98.000

92.84 - 108.32


This page is the responsibility of Jin Guojun (jguojun@lbl.gov). Support Credits are here. This document and its uses are subject to LBL's disclaimers and legal notices.

Credits: The research and development of the Distributed Systems Department is funded by the U.S. Dept. of Energy, Office of Science, Office of Advanced Scientific Computing Research, Mathematical, Information, and Computational Sciences Division

If there is a problem with this page please, e-mail webmaster-george@george.lbl.gov.


NCS HOME PAGE
DIDC HOME PAGE
LBL HOME PAGE


This page last modified: Friday, 06-May-2005 13:39:03 PDT
([an error occurred while processing this directive])