Upload
smartenit
View
105
Download
3
Tags:
Embed Size (px)
Citation preview
Traffic Profiles & Mgnt.
for Community Networks
Measurement on Network Links
– Packet and flow based analysis methods
– Traffic profiles for some large community networks
Traffic Management for Content and Service Delivery
Conclusions and Outlook
Traffic Profiles and Management
for Support of Community Networks
Gerhard Haßlinger1, Anne Schwahn2, Franz Hartleb2 1Deutsche Telekom Technik, 2T-Systems, Darmstadt, Germany
Traffic Profiles & Mgnt.
for Community Networks
Measurement of Application and Traffic Profiles
Probes can capture each IP packet: header, payload, time stamp
DPI: Content inspection (not applied for our statistics)
Analysis traffic pattern of per IP flow
A flow is identified by IP address/TCP port of source/receiver
Flow statistics are relevant for quality management
– Dimensioning with regard to variability and QoS demands
Traffic profiles are used to identify portions of applications
– We consider portions of Facebook, Twitter, Uploaded,
YouTube, VoIP
– Measurement from March’13 on 3 x 1Gb/s aggregation links
Traffic Profiles & Mgnt.
for Community Networks
Overall Measurement Statistics and Mean Values
Traffic profiles
Number of packets
[x 1000]
Packet size [Byte]
(Mean)
Number of flows
Flow size [MB]
(Mean)
Flow rate [Mbit/s]
(Mean)
Flow duration
[s] (Mean)
YouTube 17 809 1468 8 419 3.07 1.44 66
Twitter 857 662 318 0.25 0.04 129
Facebook 14 619 564 13 555 0.38 0.06 657
Uploaded 15 013 1508 508 44.54 0.46 872
Voice 4 149 295 270 4.53 0.08 455
Total traffic 1 446 065 1177 697 786 2.26 1.40 56
Traffic Profiles & Mgnt.
for Community Networks
Flow Rates for Different Application Types
Traffic Profiles & Mgnt.
for Community Networks
Flow Volume for Different Application Types
Traffic Profiles & Mgnt.
for Community Networks
Flow Durations for Different Application Types
Traffic Profiles & Mgnt.
for Community Networks
Round Trip Delays for Different Application Types
0%
20%
40%
60%
80%
100%
0,01 0,1 1
TCP Round Trip Time [s]
Total traffic
Youtube
Uploaded
Traffic Profiles & Mgnt.
for Community Networks
Traffic in Multiple Time Scales: 2nd Order Statistics
500
600
700
800
900
1000
0 1 2 3 4 5 6 7 8 9 10
Seconds
Tra
ffic
ra
te p
er
0.0
1s
in
terv
al
[Mb
it/s
]
500
600
700
800
900
1000
0 10 20 30 40 50 60
Seconds
Tra
ffic
ra
te p
er
0.1
s i
nte
rva
l [M
bit
/s]
500
600
700
800
900
1000
0 10 20 30 40 50 60
Seconds
Tra
ffic
ra
te p
er
1s
in
terv
al
[Mb
it/s
]
Evaluation of a traffic trace in 0.01s , 0.1s and 1s intervals on broadband access platform:
Variability is decreasing on larger time scales, although long range dependency persists
Traffic Profiles & Mgnt.
for Community Networks
2nd Order Statistics for Different Application Types
Traffic Profiles & Mgnt.
for Community Networks
Users
Global
Internet Access
Network
ISP
Backbone
Peering
Other
ISPs
Long paths for P2P data exchange P2P
Short CDN paths
CDN
PoPs
Points of Presence Access Control
P2P
Users
Global Content Delivery: CDN Peer-to-peer overlays
Traffic Profiles & Mgnt.
for Community Networks
Cacheability on the Internet
An essential portion of IP traffic uses HTTP protocol (80% in 2013),
most of which is marked as being cacheable, often with expiry date
Requests focus on most popular content small caches are efficient
Zipf law 90 10 rule: 90% of requests address only 10% of content
Some content providers/CDNs support caching, e.g. software updates
… others don’t: Personalised communication with user
makes content identification difficult for cache manager;
no standard feedback & control between cache content provider
Some content providers/CDNs have business relations
with content owners and/or users but often
without involving network providers
Traffic Profiles & Mgnt.
for Community Networks
IETF Standardization Groups on CDNI and ALTO
Caching is applied in global content delivery networks
and in network provider platforms of large ISPs …
but usually without much cooperation!
Content and CDN provider would like full control on client-server
activity ISP would like full control of their network and caches
IETF working group on CDN interconnection (CDNI) since 2011 <http://datatracker.ietf.org/wg/cdni/charter/>
IETF WG on Application Layer Traffic Optimization (ALTO)
- Focus on localized data exchange for P2P and other applications - ALTO servers collect data on locations of peers/clients and make it available to applications/overlay networks - Infos: provider network (AS) of endpoints; topology & cost maps - Network providers can host ALTO servers to recommend sources for content delivery without revealing their network
Traffic Profiles & Mgnt.
for Community Networks
Conclusions and Outlook
We analyzed traffic profiles of popular applications
in community networks
IP flow and packet analysis is useful for classifying portions of application traffic even without DPI
Characteristics of flow rates, volume, duration and 2nd order stat. differ for each application; community networks generate a mix of applications
For further study: QoS Characteristics in TCP round trip delay and packet loss; improved identification using traffic profiles
Popular global communities with high traffic demand are using CDN and P2P overlays, which are subject to long transport paths
Traffic optimization is considered by IETF working groups CDNI and ALTO based on cooperative approaches between administrative domains to improve local data exchange