sprobe: another tool for measuring bottleneck link bandwidth stefan saroiu p. krishna gummadi steven...

24
SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Upload: felicia-allen

Post on 16-Jan-2016

224 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe: Another Tool for Measuring Bottleneck Link

Bandwidth

Stefan SaroiuP. Krishna Gummadi

Steven GribbleUniversity of Washington

Page 2: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Peer-to-Peer File-Sharing Systems

• Ongoing project - analyze and measure characteristics of current P2P systems

• Estimate bottleneck bandwidths to and from many and diverse Internet hosts:– Traditional modems: ppp35.bcpl.net– Cable modems: c589502-

a.plano1.tx.home.com– DSL: adsl-78-148-152.cae.bellsouth.net– Dorms (10/100 Mbps):

st14_032.dorm.depaul.edu

Page 3: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Wanted: Sexy Linux Tool that:

1. Measures a network physical property - bottleneck bandwidths

2. Has reasonable accuracy• Willing to tolerate factor-of-two errors

3. Preferably returns only one estimate4. Runs fast

• 1 min/host is too slow

5. Measurement software is deployed at only one endpoint

Page 4: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Problems with Contenders

• Measuring ALL links bandwidths is slow:– pathchar, pchar, clink

• Measurement software must be deployed at both endpoints:– pathrate

• Estimates based on ICMP:– ICMP can be filtered, dropped, answered by

routers, rate-limited answers– bprobe

Page 5: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Problems with Contenders (2)

• Estimates based on passively monitoring traffic to and from the remote host:– This works as long as the traffic is formed by

large packets-pairs– Unfortunately, most of our traffic is not– nettimer

• Traffic is processed as a set of packet-pairs only:– Traffic can be shaped to reveal more

information – cross-traffic detection

Page 6: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 7: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 8: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 9: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 10: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 11: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 12: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 13: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 14: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 15: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 16: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 17: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 18: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From local host To remote host– No cooperation needed

Local Remote

SYN packet

RST packet

Page 19: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

SProbe Uses TCP Tricks

• From remote To local– Involuntary cooperation of application

layer

Local Remote

(Web)

HTTP Get request

Data packet

ACK (last data packet)

Page 20: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Trials (local to remote)

ROPP SBPP ROPP (reverse)

volkswagen.cs.washington.edu 100Mb 32Mb 32Mb 32Mb 97Mb 83Mbwww.cs.stanford.edu 100Mb 32Mb 106b 32Mb 26Mb 24MbRAS2-p218.jlm.netvision.net.il ? 32Mb 70b 19.9Kb-32Mb timeout unknownwww.cs.uwaterloo.ca 10Mb 32Mb 32Mb 2.6Mb icmp filter 4.5Mbcrtntx1-ar3-089-007.dsl.gtei.net 128Kb-2Mb 32Mb 30b 83Kb timeout 704Kbxxx.ktchnr1.on.wave.home.com 0.5-1.5Mb 32Mb 32Mb 26Kb-1.8Mb timeout 400Kbmp-216-53-118.daxnet.no ? 74Mb 32Mb 52Kb timeout 63Kbppp35.bcpl.net 14.4-56Kb 32Mb 60b 0.35-65Kb timeout unknownPPPa96-xxx.dialinx.net 14.4-56Kb 78Mb 32Mb 32Kb timeout 40Kb

sprobenettimerNominal bprobe

SProbe sends train of 6 packets (3160 bytes/train),

upto 10 times

Page 21: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Trials (local to remote)

ROPP SBPP ROPP (reverse)

volkswagen.cs.washington.edu 100Mb 32Mb 32Mb 32Mb 97Mb 83Mbwww.cs.stanford.edu 100Mb 32Mb 106b 32Mb 26Mb 24MbRAS2-p218.jlm.netvision.net.il ? 32Mb 70b 19.9Kb-32Mb timeout unknownwww.cs.uwaterloo.ca 10Mb 32Mb 32Mb 2.6Mb icmp filter 4.5Mbcrtntx1-ar3-089-007.dsl.gtei.net 128Kb-2Mb 32Mb 30b 83Kb timeout 704Kbxxx.ktchnr1.on.wave.home.com 0.5-1.5Mb 32Mb 32Mb 26Kb-1.8Mb timeout 400Kbmp-216-53-118.daxnet.no ? 74Mb 32Mb 52Kb timeout 63Kbppp35.bcpl.net 14.4-56Kb 32Mb 60b 0.35-65Kb timeout unknownPPPa96-xxx.dialinx.net 14.4-56Kb 78Mb 32Mb 32Kb timeout 40Kb

sprobenettimerNominal bprobe

SProbe sends train of 6 packets (3160 bytes/train),

upto 10 times

Page 22: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Trials (local to remote)

ROPP SBPP ROPP (reverse)

volkswagen.cs.washington.edu 100Mb 32Mb 32Mb 32Mb 97Mb 83Mbwww.cs.stanford.edu 100Mb 32Mb 106b 32Mb 26Mb 24MbRAS2-p218.jlm.netvision.net.il ? 32Mb 70b 19.9Kb-32Mb timeout unknownwww.cs.uwaterloo.ca 10Mb 32Mb 32Mb 2.6Mb icmp filter 4.5Mbcrtntx1-ar3-089-007.dsl.gtei.net 128Kb-2Mb 32Mb 30b 83Kb timeout 704Kbxxx.ktchnr1.on.wave.home.com 0.5-1.5Mb 32Mb 32Mb 26Kb-1.8Mb timeout 400Kbmp-216-53-118.daxnet.no ? 74Mb 32Mb 52Kb timeout 63Kbppp35.bcpl.net 14.4-56Kb 32Mb 60b 0.35-65Kb timeout unknownPPPa96-xxx.dialinx.net 14.4-56Kb 78Mb 32Mb 32Kb timeout 40Kb

sprobenettimerNominal bprobe

SProbe sends train of 6 packets (3160 bytes/train),

upto 10 times

Page 23: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Trials (local to remote)

ROPP SBPP ROPP (reverse)

volkswagen.cs.washington.edu 100Mb 32Mb 32Mb 32Mb 97Mb 83Mbwww.cs.stanford.edu 100Mb 32Mb 106b 32Mb 26Mb 24MbRAS2-p218.jlm.netvision.net.il ? 32Mb 70b 19.9Kb-32Mb timeout unknownwww.cs.uwaterloo.ca 10Mb 32Mb 32Mb 2.6Mb icmp filter 4.5Mbcrtntx1-ar3-089-007.dsl.gtei.net 128Kb-2Mb 32Mb 30b 83Kb timeout 704Kbxxx.ktchnr1.on.wave.home.com 0.5-1.5Mb 32Mb 32Mb 26Kb-1.8Mb timeout 400Kbmp-216-53-118.daxnet.no ? 74Mb 32Mb 52Kb timeout 63Kbppp35.bcpl.net 14.4-56Kb 32Mb 60b 0.35-65Kb timeout unknownPPPa96-xxx.dialinx.net 14.4-56Kb 78Mb 32Mb 32Kb timeout 40Kb

sprobenettimerNominal bprobe

SProbe sends train of 6 packets (3160 bytes/train),

upto 10 times

Page 24: SProbe: Another Tool for Measuring Bottleneck Link Bandwidth Stefan Saroiu P. Krishna Gummadi Steven Gribble University of Washington

Final Remarks• A useful tool (for us) needs to:

– Be deployed at one endpoint only– Run fast– Actively send probing packets– Detect cross-traffic interference

• When cross-traffic detected return ‘Unknown Estimate’• Do not returns bogus estimate instead, please.

• SProbe source will be available shortly:– http://www.cs.washington.edu/homes/tzoompy/

sprobe