growth analysis of a large isp - brown university
TRANSCRIPT
2
Cogent Communications
http://www.cogentco.com/en/network/network-map
1. One of the world’s largest IP networks, covering 3 continents2. Public map (below) provides a static snapshot at the city-level3. Since Jan. 2012, we made weekly snapshots at the router interface-level
4
Inferred Router Growth
3800 3900 4000 4100 4200 4300 4400 4500 4600 4700 4800 4900
F M A M J J A S O N D J2013
F M A M J J A S
Num
ber o
f Rou
ters
Week
5
Inferred Router Growth
3800 3900 4000 4100 4200 4300 4400 4500 4600 4700 4800 4900
F M A M J J A S O N D J2013
F M A M J J A S
Num
ber o
f Rou
ters
Week
6
Inferred Router Growth
3800
3900
4000
4100
4200
4300
4400
4500
4600
4700
4800
4900
F M A M J J A S O N D J2013
F M A M J J A S
Num
ber
of R
oute
rs
Week
r = 10.265w + 3873.7
R2=0.986
7
Interface Growth
0
10
20
30
40
50
60
J2012
F M A M J J A S O N D J2013
F M A M J J A S O
Count (x
1000)
Week
Physical Virtual
8
Physical Interface Breakdown
0 2 4 6 8
10 12 14
J2012
F M A M J J A S O N D J2013
F M A M J J A S O
Count (x
1000)
Week
FastEth10GigE1GigE
SerialEthISM
POS
9
Visualization of Inferred Paths
SDU��XUR��
IUD��EUT��
NUN��
DPV��DPV��
ORQ��
ORQ��
DPV��
PDG��PDG��
FOH��SLW��
ORQ��
MIN��MIN��
PSO��PUV��
PDG��PDG��
KDP�� ZDZ��
QFH��
GHQ��
VOF��
PH[��PIH��
DPV��
IUD��IUD��
EXK�� KNY��
DXV��
DXV��
JQE��JYD��
]UK��
OD[��OD[��
FYJ��RUG��
GXE��
GXE��
]UK��
RVO��RVO��
EXI��
\PT��
WSD��WSD��
GHQ��
PNH��
RUG��
DQU��
EUX��
XGQ��YFH��
PLO��
PXF��
ODV��ODV��
DWO��
GFD��ELR��UMO��
GFD��
EKP��MDQ��
OD[��OD[��
VMF��
LDK��
OKU��
ORQ��
IUD��
DPV��
MIN��MIN��
ORQ��
UWP��FEJ��
OED��
NES��RGV��
LDG��
LDG��
SK[��
VDQ��
FOH��
ERG��SLV��
SDU��VWR��
VIR��
VMF��
RDN��
\PT��
HOS��
VPI��
DEF��
YOF��
SDU��
SDU��
KUN��
RUG��
FOW��
VEQ��
LDK��
]DY��
EFQ��
PDQ��
PDQ��
MIN��
MIN��
FQG��
LQG��
EXK��
SDU��SDU��
FIU��UQV��
PVS��
VIR��
QWH��
OLV��
YJR��
GFD��
TOV��
EWV��EWV��
DET��DET��
ODV��
GIZ��GIZ��
VWR��
VWR��
GIZ��GIZ��
KHO��KHO��
VRI�� VRI��
VMF��
SK[��
LDG��
RNF��WXO��
UWP��
GD\��
YQR��
SG[��
RYG��
PDQ��
PLD��PLD��
PFR��
VWO��VWO��
PXF��
PNH��
O\V��
FSK��EWV��
ORQ��
EZL��EZL��
VMF��
GXV��
PFL��OSO��
V[E��
VWU��
OX[��
ERV��
RUK��
VDQ��
JRD��URP��
GFD��
LDG��
ORQ��
SDU��
UGX��
DXV��
V[E��
ORQ��
VMF��DWO��
LDK��
SYG��
EHJ�� DWK��
GIZ��
EXG��EXG��
PW\��TUR��
GFD��LDG��
]DJ��]DJ��
EHU��
GUV��
VWO��
DOE��KSQ��KSQ��
V\U��
OUG��
YLH��SDU��
PP[��
SDU��
GWZ��GWZ��
EUQ��
DWO��
GVP��RPD��
VQD��
OMX��OMX��
GIZ��
LVW��ORQ��
VMF��
VDW��
\KP��
\\]��
DJS��
VYT��
VDQ��
\PT��
FRV��
UKH��
OD[��
DPV��
SDU��SDU��
ERL��
HZU��HZU��
GWP��HVV��
PDG��
RUG��
VDQ��
ERV��
VNJ��OD[��
HZU��
PHP��
VQD��
GHQ��
SR]��
VPI��
KT��
GIZ��
VHD��
VMF��
EVO��
ORQ��ORQ��
WOV��
FOH��
OD[��
EUH��
DWO��
JVR��
SKO��SKO��
NLY��
RUG��
HGL��
GHQ��
ORQ��MD[��
SKO��
PLD��PLD��
EFW��EFW��
VDQ��OD[��
GXE��
JU[��
GFD��
FES��
VMF��
W\R��
LDG�� DPV��VHD��RUG��
VHD��VHD��
MIN��
GHY��
MIN��
SG[��
PLD��
ORQ��
WXI��
DPV��
GHY��
IUD��
WOO��
\\]��
T]R��
\\]��
SUJ��
RUG��
EUX��
KHO��
ORQ��ORQ��
DXV��
FPK��
ZDZ��
]UK��
GHQ��
SDU��
HDV��
SK[�� PLD��
PLD��
V\U��
GFD��
\YU��
MIN��
VMF��
PDQ��MIN��
HZU��
PLD��
VIR��
EXI��
OLO��
ODV��
LDK��
GD\��
VQD��
FYJ��
IUD��IUD��
PLD��
EM]��
DWO��
IUD��
OMX��
IUD��
VWR��
ORQ��WVU��
SDU��
VRI��
QFH��LDG��
DXV��
PMY��
]UK��
ORQ��
VGI��
VMF��
PDG��
FOW��
SK[��
VDQ��
IWZ��
HQV��
V[E��
OD[��
YOF��
PLD��
NES��
IUI��
\PT��
LDG��
GIZ��
GIZ��
QXH��
MIN��
RUG��
LDG��
GXV��
IOU��
DWO��PV\��
RDN��
OD[��
RUG��
ERV��
ORQ��
VMF��
JOD��
RSR��WRO��
PH[��
SKO��
VNS��
ERV��
OD[��
IPR��
JGO��
DPV��
SK[��
VOF��
GHY��
LDG��
HZU��MIN��
LDK��
VHD��
GHY��
DWO��
1. Infer connection between two routers sharing appropriate /30 subnets2. Nodes are sized according to the number of paths passing through them3. Layout above is force-directed (no geographical information used)
11
Cogent’s DNS Records$ host 154.54.80.85te2-1.ccr01.jfk01.atlas.cogentco.com
Ten GigabitEthernet
Slot 2
Port 1
RouterInfrastructure
Metro
11
Cogent’s DNS Records$ host 154.54.80.85te2-1.ccr01.jfk01.atlas.cogentco.com
$ host 154.54.25.17te2-2.ccr01.jfk01.atlas.cogentco.com
Ten GigabitEthernet
Slot 2
Port 1
RouterInfrastructure
Metro
12
Cogent’s DNS Records (2)$ host 38.112.5.17fa0-2.na01.b003070-1.sfo04.atlas.cogentco.com
100 MbpsEthernet
Router Metro
12
Cogent’s DNS Records (2)$ host 38.112.5.17fa0-2.na01.b003070-1.sfo04.atlas.cogentco.com
$ host 38.112.5.18Tetratech.demarc.cogentco.com
100 MbpsEthernet
Router Metro
12
Cogent’s DNS Records (2)$ host 38.112.5.17fa0-2.na01.b003070-1.sfo04.atlas.cogentco.com
$ host 38.112.5.18Tetratech.demarc.cogentco.com
100 MbpsEthernet
Router Metro
Pair in /30 Subnet
Related businessentities
Californiaengineering firm
13
Weekly Surveys1. Perform 20+ million reverse DNS queries weekly for Cogent-owned IPs2. Issued from ~100 PlanetLab locations across the globe3. Also run iffinder on the previous week’s discovered interfaces (~55k)
0 2 4 6 8
10 12 14 16 18 20 22
J2012
F M A M J J A S O N D J2013
F M A M J J A S
Nu
mb
er
of
Pro
be
s (x
10
6)
Week
Other ErrorsServer Failure
Timeout
No Reverse MappingSuccessful
14
Weekly Surveys1. Perform 20+ million reverse DNS queries weekly for Cogent-owned IPs2. Issued from ~100 PlanetLab locations across the globe3. Also run iffinder on the previous week’s discovered interfaces (~55k)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
J2012
F M A M J J A S O N D J2013
F M A M J J A SNum
ber
of P
robes
(x10
6)
Week
Non-Cogent Cogent-Other Cogent Atlas
16
2. Check Cogent’s public information1. Compare with iffinder
3. Use complete set of IPv4 DNS records
Validation Approaches
17
Comparison with iffinderiffinder — a well-known solution to the “alias resolution” problem with a low rate of false positives
18
Comparison with iffinderiffinder — a well-known solution to the “alias resolution” problem with a low rate of false positives
ccr01.jfk01 ccr02.jfk01
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
19
Comparison with iffinderiffinder — a well-known solution to the “alias resolution” problem with a low rate of false positives
ccr01.jfk01 ccr02.jfk01
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
Good!
20
Comparison with iffinderiffinder — a well-known solution to the “alias resolution” problem with a low rate of false positives
ccr01.jfk01 ccr02.jfk01
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
te2-
1
te2-
2
te2-
3
te2-
4
te2-
5
te2-
6
te2-
7
te2-
8
Bad!
22
Andrew [email protected]
• Jordan Place
• Rodrigo Fonseca
Co-a
utho
rs
systems.cs.brown.edu/cogent