test results prof. timothy a. gonsalves i.i.t. madras
TRANSCRIPT
![Page 1: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/1.jpg)
Broadband Quality of Service ExperienceQoSE
Test Results
Prof. Timothy A. GonsalvesI.I.T. Madras
3 Nov 09
![Page 2: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/2.jpg)
Testing Rounds
• Mar 08: India, Sri Lanka• Oct 08: India, Sri Lanka, Bangladesh• Mar 09: India, Sri Lanka, Bangladesh• Oct 09: India, Sri Lanka, Bangladesh, US,
Canada
This work was carried out with the aid of a grant from theInternational Development Research Centre, Canada and theDepartment for International Development, UK.
![Page 3: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/3.jpg)
Test Methodology• Tests 6 parameters
– Covers most aspects of subscriber QoE
• Tests three servers (ISP, National,International)– ISP local network, National Peering and
International Bandwidth
• Repeated at different times of the day– Provisioning & Dimensioning of Network
• Repeated at weekdays and weekends– Provisioning & Dimensioning of Network
ISP National International
Tim
es o
f Day
Wee
kday
s
Wee
kend
s
Automated by the ATTester http://www.broadbandasia.info
![Page 4: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/4.jpg)
Tests Done in Oct 09
• Locations:– Delhi, Mumbai, Pune, Bangalore, Chennai– Tamilnadu
• Operators:– Airtel, BSNL, TATA, MTNL
• Packages:– Wired: 256 kb/s - 2 Mb/s, limited/unlimited– Wireless: 256 kb/s, limited (TN)
• Globally:– US: Verizon (3 Mb/s), Comcast (6 Mb/s)– Canada: Bell (6 Mb/s), Rogers (10 Mb/s)
Using the ATTester http://www.broadbandasia.info
![Page 5: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/5.jpg)
Test Results Download speed, RTT, Jitter:
• Sample results
• All India metros: National/Int’l domains
• India vs. N. America
RTT Analysis
BSNL
• Limited vs. unlimited
• Tamilnadu vs. Chennai
• Wired vs. wireless
![Page 6: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/6.jpg)
Sample Test Results
Download speed, RTT, jitter,loss metrics
![Page 7: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/7.jpg)
Download speed (Chennai/Delhi 256k unlimited):Relatively healthy in National domains, Int’l much lower
• Nat’l good
• Int’l marginal
• Both operatorssimilar
Chennai BSNL
Delhi Airtel
![Page 8: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/8.jpg)
Round-trip time (RTT) (256k unlimited):TRAI: <350 ms, Singapore <300 ms VOIP needs <100 ms (one way)
• Nat’l good
• Int’l marginal
• Both operatorssimilar
Chennai BSNL
Delhi Airtel
![Page 9: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/9.jpg)
Jitter (256k unlimited):<10 ms very good, peaks >50 ms affect VOIP
• Jitter mostly good
• Some randomvariation
• BSNL poorer as ituses Airtel Int’lgateway?
Chennai BSNL
Delhi Airtel
![Page 10: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/10.jpg)
Where is the Bottleneck?
![Page 11: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/11.jpg)
Nat’l Round-trip time -- hop-by-hopTrace route from PC in Chennai to server
• Gateway to otherISP is bottleneck
• Airtel: ADSL linkcauses high RTT
• RTT 10-15 msachieveable
Chennai BSNL
Delhi Airtel
![Page 12: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/12.jpg)
Int’l Round-trip time -- hop-by-hopTrace route from PC in Chennai to server
• Int’l gateway isbottleneck
• RTT 100-200 msachieveable
RTT to cpan.pair.com (global address)
1
38 30 30
284
320 326
251
0
100
200
300
Local-Modem
airtel-Chennai
airtel-Chennai
airtel-Delhi
airtel-Bangalore
GlobalCrossing-US
GlobalCrossing-Penn
PairNetworks-Penn
Route in sequence
RT
T
in m
s
Chennai BSNL
Delhi Airtel
![Page 13: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/13.jpg)
Using RTT/latency in broadband QoSregulation
National Network Latency (NNL): Round Trip Time (RTT)taken to access a national or site hosted within thecountry
Requirement: NNL < 85ms for 95% of the peak time
International Network Latency (INL) : Round Trip Time(RTT) to the first overseas entry point
Requirement: INL < 300ms for 95% of the peak time
![Page 14: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/14.jpg)
Analysis of RTT
o Apparently unnecessary hops between cities
o National gateway between operators is a bottleneck
o Int’l gateway is the main bottleneck
o RTT is key because of TCP behaviour
o Delay is due to link bandwidth, and router congestion
o TCP adapts speed to end-to-end metrics RTT and loss
!
Speed"1
RTT#
1
Loss
![Page 15: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/15.jpg)
Sample Results 256k unlimited: Points to ponder
o ISPs give much better than minimum advertised ratein ISP domain, but most users don’t use ISP servers
o National download is much better than minimum, RTTmeets international standards (<85 ms)
o Int’l download often does not meet minimum, RTT ishigher than int’l/TRAI norms
==> Need tuning of gateway + more int’l bandwidth==> Need more caching/mirroring/hosting of content inIndia
![Page 16: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/16.jpg)
All-India Metro Results
![Page 17: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/17.jpg)
National Download speed (all India metro):Relatively healthy in ISP/National domains…
• Bangalore is BSNL gateway to other operators, soBLR-BSNL has good speed• Delhi is Airtel gateway city• Most operators “average performance”
![Page 18: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/18.jpg)
Int’l download speed (all India metro):
• Curiosity: BSNL did better than Airtel although BSNLuses Airtel for int’l connectivity• QoSE of different plans not necessarily related to cost
![Page 19: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/19.jpg)
Int’l download/minimum speed % (all Indiametro):
• Most plans under 100%• BSNL gave better than minimum
![Page 20: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/20.jpg)
All India metro: Points to ponder
o National download speed is well above the planminimum in most cases
o Int’l download speed barely reaches the plan minimumin most cases
o Subscribers in their operator’s “gateway city” seem toget better QoSE
o Caveat: comparative results are only indicative, needmany more tests for each operator to draw definiteconclusions
![Page 21: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/21.jpg)
India vs. N. America
Throughput and Delay
![Page 22: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/22.jpg)
Download speed (India and US):India: ISP good, big differences; US: only small differences
Verizon NY
Delhi Airtel
![Page 23: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/23.jpg)
RTT (India):ISP/Nat’l: very good; Int’l: barely acceptable
Chennai BSNL
Delhi Airtel
![Page 24: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/24.jpg)
RTT (US):ISP/Nat’l: very good; Int’l: good
Verizon NY
Comcast CO
![Page 25: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/25.jpg)
India vs. N. America: Points to ponder
o India
• considerable variation across time-of-day
• very high difference between national and int’l
o N. America
• uniformly good QoSE almost always -- due to over-provisioning and good network management
![Page 26: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/26.jpg)
Analysis of BSNL Plans
BSNL selected because of larger number of wired broadbandsubscribers
![Page 27: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/27.jpg)
BSNL 256k limited vs. unlimitedDownload speed
Unlimited gave lower QoSE than limited:apparently less value for more money
Of course, Unlimited gives more monthly download
![Page 28: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/28.jpg)
Premium plans (BSNL Tamilnadu)QoSE may not increase with “higher” plans
• Last-mile is not the bottleneck,so increasing minimum in it doesnot help
• End-to-end speed of TCPdepends on RTT and loss (notraw bandwidth)
• Premium plans offer otherbenefits e.g. higher data transfer
256K
512K
![Page 29: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/29.jpg)
Chennai vs. Tamilnadu (256k limited):The digital divide still exists
Also, power failurescommon in rural areas
![Page 30: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/30.jpg)
BSNL TN Mobile 256K limited (one location)Download (Kbps) 8:00 AM 11:00AM 03:00PMISP 254 216 183National 353 194 174Int'l 317 167 136
RTT (ms) 8:00 AM 11:00AM 03:00PMISP 436 890 749National 387 522 570Int'l 644 576 628
Jitter (ms) 8:00 AM 11:00AM 03:00PMISP 327 613 600National 235 462 460Int'l 340 289 344
High RTT and very high jitter for all domains -- results in similar downloadspeed for all domains -- does not meet nat’l/int’l standards
![Page 31: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/31.jpg)
Points to ponder
o Paying more money does not imply better QoSE --customer needs more technical awareness
o Rural urban divide still exists, could be reduced
o Wireless broadband may be very different from wiredbroadband
![Page 32: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/32.jpg)
Other Drivers for Broadband
![Page 33: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/33.jpg)
InfrastructureBroadband costs are low: Rs. 2.5K-12K/year
Connectivity is rapidly proliferating1. Power is very unreliable in rural areas2. Needs a PC: Rs. 20K replaced every 3
years– Plus software, maintenance hassle
Solution: low-cost thin client such asNovatium’s NetPC and Nova Navigator– Maintenance-free, low power appliance– Computing as a utility service, monthly cost
similar to broadband tariff
![Page 34: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/34.jpg)
Conclusions
o All operators offer broadly similar QoSE
o Over-advertising is rare (thanks to TRAI and operators)but subscribers may misunderstand
o Routing of packets esp. between operators can beimproved -- bigger role for NIXI?
o International QoSE is a bottleneck across the board --scope for fine-tuning gateway parameters
o Definition of broadband should shift away from rawlast-mile speed -- specify QoSE to reference servers?
o Infrastructure (power) and low-cost, reliable accessterminals are key to expansion of rural broadband (70%pop.)
![Page 35: Test Results Prof. Timothy A. Gonsalves I.I.T. Madras](https://reader031.vdocuments.mx/reader031/viewer/2022012916/61c6562f47680b7e8b07567c/html5/thumbnails/35.jpg)
Thank You!
[email protected] [email protected] www.lirneasia.net
Thanks to the LIRNEasia, TeNeT, Nilgiri Networks and DesiCrew teams,and to R. Thirumurthy and C. Jagadish
This work was carried out with the aid of a grant from theInternational Development Research Centre, Canada and theDepartment for International Development, UK.