![[CAIDA - Center for Applied Internet Data Analysis logo]](/images/caida_globe_faded.png "Go to CAIDA home page")
Goals and objectives
GOAL: Compare and contrast different bandwidth estimation tools according to the same metrics.
Currently available bandwidth estimation tools measure different bandwidth metrics, use different methodologies and generate different amounts of overhead traffic. We wish to identify the strengths and weaknesses of each tool in order to provide a "Consumer's Report" style analysis that will allow researchers and application developers to choose an appropriate tool to meet their needs.
OBJECTIVE: Characterize tools and prioritize order of evaluations. Note: tools in parentheses haven't yet been installed.-
Per-hop (link) capacity tools: pathchar, pipechar, pchar, clink
-
End-to-end capacity tools: Sprobe, pathrate, nettimer, (bprobe)
-
End-to-end available bandwidth tool: (pathload)
-
BTC or TCP throughput tools: (bbcp, cap)
-
Parallel TCP connections throughput tools: Iperf, (netperf)
-
Measure on known capacity links
-
Measure while increasing link utilisation in 5% steps (via Smartbits traffic generator)
-
Measure while injecting different types of cross-traffic:
- constant bitrate fixed size UDP packets (1500 B)
- random mix of 40/576/1500 B packets
- single large TCP stream
- 5/10/20 random-size file transfer streams
-
Measure/estimate UCSD traffic for 24 hr weekday on known path
-
Measure/estimate Abilene traffic for 24 hr weekday on known path
-
Read router SNMP counters in parallel with above .
Testing Methodology
Testlab configuration (simulated traffic)
- Smartbits6000 Traffic Generator configured by console PC connected to its backend control port
- Host wednesday: NeTraMet flow meter/passive monitor atttached to 100M load path via 100Mb copper tap and to GigEther load path via fiber repeater
- Cisco Catalyst 2900 switch
- Extreme Networks Summit 5i router
- Juniper M20 router
- Foundry Big Iron router
- Host bw-est-1: CAIDA 100M bwest test box A (Sender)
- Host bw-est-2: CAIDA 100M bwest test box B (Receiver)
- Host fester: CAIDA GigEther bwest test box A (sender)
- Host gomez: CAIDA GigEther bwest test box B (Receiver)
- n pairs of network interfaces managed by Smartbits:
- 10/100/1000 Ethernet
- GigEther (fiber)
- OC48 (1 pair)
Testlab configuration (real traffic, with known switches and routers)
- NeTraMet flow meter/passive monitor
- mesh of bwest and re-purposed skitter boxes:
- CAIDA bwest test box A (???)
- CAIDA bwest test box B (???)
- chenin (Boulder, CO)
- champagne (Illinois)
- zinfandel (???)
- merlot (???)
Timeline
| Task | Date | People |
|---|---|---|
| Configure Reference Lab with 10/100M NICs at SDSC | 12Mar02 | Kevin, marg, vineet |
| Run tools against simulated traffic in closed lab | Apr-Jun02 | vineet |
| Analyze and graph results | Apr-Jun02 | vineet, marina |
| Configure end-host on UCSD path(s) | Jul02 | Kevin, marg, vineet |
| Run tools from machine in CalNGI lab to UCSD end host. Track level of real cross-traffic | Jul-Aug02 | vineet,ravi |
| Analyze and graph results | Aug02 | vineet,marina |
| Configure end-host on Abilene path(s) | Aug02 | ravi |
| Run tools on Abilene path using real cross-traffic | Aug-Sep02 | vineet, ravi |
| Analyze and graph results | Sep02 | vineet, marina |
| Draft of results paper | Aug1502 | vineet, ravi, marg |
| Draft of sysadmin/netops paper | Oct02 | vineet,marg |