Analysing impact of Bangaldesh transit for North East India

This Sunday, telecom news and even some general news in India reported that Bangladesh has cancelled the bandwidth supply agreement for North East India. This would add to the poor decision-making, which has been happening in Bangladesh since 5th August. The decision-making overall seems childish and regressive.


What has happened?

In 2016, the Indian Govt started taking bandwidth for North East India which geographically was a bit far from other major bandwidth PoPs in India and close to Bangladesh. I remember writing a blog post about it back then. I see some articles state it to be 2023, though I have seen BSCCL being the upstream IP transit player of BSNL AS9289 since 2016.

As per the Economic Times India report, “the decision impacted Bangladesh’s ability to provide internet services to the parts of Myanmar and North Western China”. I honestly have no idea of Mynamnar but strongly doubt that China is looking for a bandwidth transit route via Bangladesh.


IP transit situation right now?

Let’s look at routes announced by BSNL AS9829 behind BSCCL (Bangladesh Submarine Cables Company) AS132602. There are 65 prefixes visible in the global table with AS_PATH “132602 9829”. Out of these 65, 47 are IPv4 and 18 IPv6. I have posted a detailed list here.

So this brings us to two questions - is latency significantly better and are these pools at risk of disconnection? I picked 117.198.64.1 from 117.198.64.0/20 BSNL to BSCCL, Tata Comm AS4755, Airtel AS9498, Tata Comm AS6453 (outside of India) and Cogent AS174.


Comparing routes now for latency:

Picking 117.250.113.140 - active RIPE Atlas anchor on BSNL Guwahati and IP is part of 117.250.112.0/20 announced by BSNL to BSCCL, Tata Comm AS4755 and Airtel AS9498.

Source: bgp.he.net

1. Transit via Bangladesh

Equinix Singapore > BSSCL > BSNL India

1 2.112 ms  3.189 ms  2.041 ms  132602.sgw.equinix.com (27.111.229.146)
2 1.931 ms  2.702 ms  0.796 ms  103.16.153.18 (103.16.153.18)
3 41.259 ms 41.368 ms 43.982 ms 103.16.153.26 (103.16.153.26)
4 48.329 ms 48.691 ms 53.818 ms 103.16.152.81 (103.16.152.81)
***
7 72.918 ms 72.791 ms 72.802 ms static.ill.117.250.113.140.bsnl.co.in (117.250.113.140)

Hop 2 - is BSCCL Singapore (router which connects to Equinix on one side and submarine cable on the other) Hop 3 - BSCCL Cox Bazar via SE-ME-WE 4 Hop 4 - Likely Dhaka Hop 7 - BSNL Guwahati


2. Transit via Chennai

Tata Comm AS6453 (known upstream of AS4755) Singapore > Tata Comm Chennai > Tata Comm Mumbai > BSNL Mumbai »» BSNL NLD + Powergrid OGW (likely) > BSNL NE Guwahati

Router: gin-svq-thar1
Site: SG, Singapore, SVQ
Command: traceroute inet4 117.250.113.140 as-number-lookup

traceroute to 117.250.113.140 (117.250.113.140), 30 hops max, 52 byte packets
 1  if-be-40-2.ecore1.svq-singapore.as6453.net (120.29.214.101)  1.486 ms  1.790 ms  1.562 ms
 MPLS Label=24272 CoS=0 TTL=1 S=1
 2  * * *
 3  if-bundle-19-2.qcore2.svw-singapore.as6453.net (63.243.180.64)  2.041 ms  1.334 ms *
 MPLS Label=45547 CoS=0 TTL=1 S=1
 4  if-ae-33-2.tcore2.svw-singapore.as6453.net (180.87.84.128)  1.716 ms * *
 5  180.87.15.202 (180.87.15.202)  63.783 ms  60.485 ms  61.553 ms
 6  172.31.231.230 (172.31.231.230) [AS  4761]  85.906 ms  85.896 ms  82.768 ms
 7  * * *
 8  * * *
 9  * * *
10  * * *
11  static.ill.117.250.113.140.bsnl.co.in (117.250.113.140) [AS  9829]  103.738 ms  99.899 ms  102.419 ms



Router: gin-svq-thar1
Site: SG, Singapore, SVQ
Command: ping inet count 5 117.250.113.140

PING 117.250.113.140 (117.250.113.140): 56 data bytes
64 bytes from 117.250.113.140: icmp_seq=0 ttl=58 time=106.478 ms
64 bytes from 117.250.113.140: icmp_seq=1 ttl=58 time=106.808 ms
64 bytes from 117.250.113.140: icmp_seq=2 ttl=58 time=106.459 ms
64 bytes from 117.250.113.140: icmp_seq=3 ttl=58 time=106.954 ms
64 bytes from 117.250.113.140: icmp_seq=4 ttl=58 time=106.241 ms

--- 117.250.113.140 ping statistics ---
5 packets transmitted, 5 packets received, 0% packet loss
round-trip min/avg/max/stddev = 106.241/106.588/106.954/0.257 ms

{master}

Latency comparison

Source Network Destination Path Latency
Equinix Singapore BSNL Guwahati - 117.250.113.140 via BSCCL Bangladsh 72ms
Tata Comm Singapore BSNL Guwahati - 117.250.113.140 via Chennai-Mumbai-North East 106ms
Airtel Mumbai AS24560 BSNL Guwahati - 117.250.113.140 via Mumbai-North East 58ms



Are BSNL routes behind BSCCL redundant?

Yes! Each one of the 65 prefixes is announced in the global table from other upstreams as well.


Why it may not matter?

One thing which has changed over time is the presence of content and CDN networks in India. Most of the earlier ideas were based on the logic that “content exists in Singapore” and the distance of Singapore -> Submarine Cable -> Chennai -> North East India was way more than Singapore -> Submarine Cable -> Cox Bazar, Bangladesh -> North East India. Once networks have local caches within the North East besides Indian Cloud players hosting content within India, it’s questionable if this helps routing-wise at all for the majority of traffic as measured by traffic volume. Capacity/redundancy wise it might still have helped due to hard North East terrain. It would eventually be sorted out as more dense mesh of cables connect North East Indian states with West Bengal.


Disclaimer: The views expressed on this blog are my own and do not reflect those of my employer.


Update - 10 Dec 2024
My friend Sunil Tagare posted about this topic on Linkedin. Comments in his post reply point to a different case than BSCCL - BSNL arrangement. There is some pointer to towards CDN players out of Kolkata serving more traffic to BD while BD trying to push CDN players to be present inside BD. I am unsure how true is that. However it’s important to note that few years ago BD did close down most of CDN caches sitting inside the local ISPs & allowed only IIGs to have those. Last I heard - they did allow the caches inside mobile operators as well but not the ISPs.