29 May

What makes BSNL AS9829 as most unstable ASN in the world?!

On weekend  I was looking at BGP Instability Report data. As usual (and unfortunately) BSNL tops that list. BSNL is the most unstable autonomous network in the world. In past, I have written previously about how AS9829 is the rotten IP backbone.

 
This isn’t a surprise since they keep on coming on top but I think it’s well worth a check on what exactly is causing that. So I looked into BGP tables updates published on Oregon route-views from 21st May to 27th May and pulled data specifically for AS9829. I see zero withdrawals which are very interesting. I thought there would be a lot of announcements & withdrawals as they switch transits to balance traffic.
If I plot the data, I get following chart of withdrawals against timestamp. This consists of summarised view of every 15mins and taken from 653 routing update dumps. It seems not feasible to graph data for 653 dumps, so I picked top 300. Here’s how it look like:


 
Except for few large spikes, it seems to have a relatively consistent pattern. We can see daily fresh announcements of close to 50,000 announcements.
This data gives no idea and I can’t say much by looking at it. Instead of looking at updates, I pulled last weeks RIBs and pulled AS9829 announcements. The idea here is to get map announcements to each upstream against time stamp along with announcements across various subnet masks.

Here’s total route announcement graph:
The graph above clearly shows that total routes announcements increased significantly on 23rd May at 06:00 UTC from 127664 to 129298. Thus dipped significantly at 14:00 on 26th May to 124301. So between 10:00 to 14:00 on 26th, the drop in routes as much as 4% drop clear indicating a large outage they had in their network.

Next part is to look at how they tweak their announcements to upstream.


So clearly they are announcing a large number of routes to Tata AS6453 and these are IPLC links where they are buying IP transit outside India. Some of these key spikes show a mirror among other transit giving a clear hint of circuit balancing by moving route announcement.
 
Next part is to view their announcements in terms of prefix size.



/20 as well as /22 as both seems relatively consistent except showing a dip on 26th.
 
So all I can say based on above data is following:

  1. BSNL had some issues last week. Possibly one of their upstream pipes had issues and they increased their announcements on Tata AS6453 during that time.
  2. They are an only large operator who is buying transits from as many as 9 upstream. This would result in broken capacity across at least 9 and possibly 30-40 circuits resulting in a major capacity management challenge across these upstream.
  3. They are announcing a large number of prefix sizes. /18, /20, /22, /23 and even /24s. This isn’t good practice at their large scale.
  4. They need to start peering. They are the only network of that scale who isn’t peering except with a couple of content players like Google AS15169. They need to peer aggressively inside India & follow same outside India if they actually keep on running such network. Or else even buying transit domestic only will be a better strategy.

 
Most of these problems can be fixed if BSNL aggregates it’s a number of transits (and circuits per transit) along with aggregation of routes. For a three transit scenario, they can follow /18, /20 and /22 strategy and leave /24 only for emergency cases to balance traffic.

01 Jun

BSNL > Softlayer connectivity problem & possible fix

It’s late night here in India. I am having final 8th semester exams and as usual really bored! 

Though this time we have interesting subjects but still syllabus is pretty boring spreading across multiple books, notes and pdf’s. Anyways I will be out of college after June which sounds good.

 

Tonight, I found a routing glitch. Yes a routing glitch!! 🙂

These issues somehow keep my life in orbit and give a good understanding on how routing works over the Internet.

 

 

OK – so the issue

I noticed a really bad (forward) route from my BSNL’s connection to hostgator.in website hosted in Softlayer Singapore. Let’s look at forward path:

anurag:~ anurag$ traceroute -a hostgator.in
traceroute to hostgator.in (216.12.194.67), 64 hops max, 52 byte packets
1 [AS65534] router.home (10.10.0.1) 1.189 ms 0.910 ms 0.810 ms
2 [AS9829] 117.220.160.1 (117.220.160.1) 17.707 ms 21.147 ms 16.925 ms
3 [AS9829] 218.248.169.126 (218.248.169.126) 30.195 ms 29.766 ms 29.976 ms
4 [AS9829] 218.248.250.82 (218.248.250.82) 75.432 ms 77.488 ms 76.761 ms
5 [AS6453] if-11-1-1.mcore3.laa-losangeles.as6453.net (209.58.85.5) 368.104 ms 303.206 ms 309.964 ms
6 [AS6453] if-10-2-0-14.tcore2.lvw-losangeles.as6453.net (216.6.84.6) 309.070 ms 308.725 ms 310.073 ms
7 [AS6453] 216.6.84.66 (216.6.84.66) 317.050 ms 318.714 ms 398.408 ms
8 [AS2914] ae-5.r21.lsanca03.us.bb.gin.ntt.net (129.250.5.85) 305.672 ms * 304.480 ms
9 [AS2914] as-2.r20.osakjp01.jp.bb.gin.ntt.net (129.250.3.202) 414.205 ms
[AS2914] as-1.r21.tokyjp01.jp.bb.gin.ntt.net (129.250.3.146) 485.451 ms
[AS2914] as-2.r20.osakjp01.jp.bb.gin.ntt.net (129.250.3.202) 414.272 ms
10 [AS2914] ae-3.r24.tokyjp05.jp.bb.gin.ntt.net (129.250.6.188) 381.221 ms
[AS2914] ae-1.r23.osakjp01.jp.bb.gin.ntt.net (129.250.2.49) 420.412 ms
[AS2914] ae-3.r25.tokyjp05.jp.bb.gin.ntt.net (129.250.6.192) 372.768 ms
11 [AS2914] ae-7.r25.tokyjp05.jp.bb.gin.ntt.net (129.250.3.223) 394.899 ms
[AS2914] ae-7.r24.tokyjp05.jp.bb.gin.ntt.net (129.250.3.221) 406.922 ms
[AS2914] ae-2.r00.tokyjp03.jp.bb.gin.ntt.net (129.250.2.5) 491.190 ms
12 [AS2914] ae-3.r00.tokyjp03.jp.bb.gin.ntt.net (129.250.4.233) 399.065 ms
[AS2914] xe-0-0-0.bbr01.eq01.tok01.networklayer.com (61.213.145.38) 307.955 ms
[AS2914] ae-2.r00.tokyjp03.jp.bb.gin.ntt.net (129.250.2.5) 392.937 ms
13 [AS2914] xe-0-0-0.bbr01.eq01.tok01.networklayer.com (61.213.145.38) 310.298 ms
[AS36351] ae1.bbr01.eq01.sng02.networklayer.com (50.97.18.165) 306.396 ms
[AS2914] xe-0-0-0.bbr01.eq01.tok01.networklayer.com (61.213.145.38) 407.191 ms
14 [AS36351] ae5.dar01.sr03.sng01.networklayer.com (50.97.18.197) 388.660 ms
[AS36351] ae5.dar02.sr03.sng01.networklayer.com (50.97.18.199) 303.546 ms 409.645 ms
15 [AS36351] po2.fcr01.sr03.sng01.networklayer.com (174.133.118.133) 407.589 ms
[AS36351] ae5.dar02.sr03.sng01.networklayer.com (50.97.18.199) 310.587 ms
[AS36351] po2.fcr01.sr03.sng01.networklayer.com (174.133.118.133) 305.969 ms
16 [AS36351] po2.fcr01.sr03.sng01.networklayer.com (174.133.118.133) 363.405 ms * 309.151 ms
17 * * *
18 * * *

 

BSNL (India) >> IPLC circuit >> Tata AS6453 Los Angeles, California >> NTT (US) >> NTT (Asia) >> NTT (Tokyo) >> Softlayer (Tokyo) >> Softlayer (Singapore)

Wow!

Pretty bad. Ideally route should be BSNL > Upstream – Tata/Reliance/Airtel/Vodafone > Singapore (that’s it. Over!)

 

Interesting enough that Softlayer operates a nice looking glass and hence I was able to trace return path to my home router from there to get idea of complete route.

bbr02.eq01.sng02> traceroute 117.220.163.128
HOST: bbr02.eq01.sng02-re0 Loss% Snt Last Avg Best Wrst StDev
1. 63.218.213.173 0.0% 5 0.4 0.5 0.4 0.5 0.0
2. 63.218.228.65 0.0% 5 0.6 0.6 0.5 0.7 0.0 <<< PCCW Global
3. 120.29.215.33 0.0% 5 11.1 7.4 0.6 12.7 5.1 <<< Tata AS6453
4. 120.29.214.13 0.0% 5 0.6 2.4 0.6 9.6 4.0
5. 180.87.12.9 0.0% 5 62.1 61.2 60.7 62.1 0.6
6. 180.87.12.54 0.0% 5 97.7 73.3 60.8 97.7 17.4
7. 180.87.36.33 0.0% 5 103.2 75.0 59.6 103.2 18.1
8. 180.87.38.74 0.0% 5 61.1 74.0 61.1 88.9 12.4 <<< Tata AS6453
9. 115.114.131.138 0.0% 5 91.7 92.6 91.7 96.3 2.0 <<<< VSNL AS4755
10. 218.248.255.101 0.0% 5 95.5 96.5 95.5 99.7 1.8 <<<< Hits BSNL AS9829
11. 218.248.169.117 0.0% 5 106.6 110.4 106.4 126.2 8.8
12. 218.248.169.117 0.0% 5 106.3 107.0 106.3 108.6 1.0
13. ???

 

 

Overall pretty good and direct. Basically latency value is also as we expect till hop 12 because forward route (i.e from BSNL > Softlayer) is direct from BSNL router on hop 12 but for routers below it they are taking route via US. Return path trace is not showing those routers because BSNL is dropping ICMP.

 

Reason for problem:

Forward path is terribly bad here because BSNL let usual BGP route selection algorithm to deal with it. Basically BSNL is getting multiple routes for that prefix from Softlayer. One from it’s IP port in India with Tata-VSNL AS4755 and other from it’s port from Tata in Los Angles (Tata AS6453) over IPLC.

 

So possible routes as per AS paths are:

AS9829 > AS4755 > AS6453 > AS2914 > AS36351 

AS9829 > AS6453 > AS2914 > AS36351

 

Based on default property of BGP, it is picking short AS path i.e 2nd one. In case of #1 BGP session between BSNL AS9829 and Tata-VSNL AS4755 is within India. 

For example:

1 [AS65534] router.home (10.10.0.1) 1.709 ms 0.912 ms 0.982 ms
2 [AS9829] 117.220.160.1 (117.220.160.1) 17.451 ms 18.075 ms 19.029 ms
3 [AS9829] 218.248.169.122 (218.248.169.122) 21.843 ms 24.584 ms 22.491 ms
4 [AS4755] 115.114.57.165.static-mumbai.vsnl.net.in (115.114.57.165) 57.399 ms 58.563 ms 57.446 ms

 

Very likely BGP session here is configured on usual /30 subnet with one IP on BSNL side, one on Tata’s side, third one as broadcast and 4th lying useless due to Math game!

So 115.114.57.165 is part of that /30. Let’s ping it:

anurag:~ anurag$ ping -c 5 115.114.57.165
PING 115.114.57.165 (115.114.57.165): 56 data bytes
64 bytes from 115.114.57.165: icmp_seq=0 ttl=58 time=63.286 ms
64 bytes from 115.114.57.165: icmp_seq=1 ttl=58 time=66.029 ms
64 bytes from 115.114.57.165: icmp_seq=2 ttl=58 time=59.063 ms
64 bytes from 115.114.57.165: icmp_seq=3 ttl=58 time=59.439 ms
64 bytes from 115.114.57.165: icmp_seq=4 ttl=58 time=61.719 ms

— 115.114.57.165 ping statistics —
5 packets transmitted, 5 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 59.063/61.907/66.029/2.573 ms
anurag:~ anurag$

 

60ms latency – for sure Mumbai and all good here.

 

Now let’s look at IP just next to it:

 

anurag:~ anurag$ ping -c 5 115.114.57.166
PING 115.114.57.166 (115.114.57.166): 56 data bytes
64 bytes from 115.114.57.166: icmp_seq=0 ttl=251 time=28.784 ms
64 bytes from 115.114.57.166: icmp_seq=1 ttl=251 time=25.586 ms
64 bytes from 115.114.57.166: icmp_seq=2 ttl=251 time=28.631 ms
64 bytes from 115.114.57.166: icmp_seq=3 ttl=251 time=26.905 ms
64 bytes from 115.114.57.166: icmp_seq=4 ttl=251 time=26.213 ms

— 115.114.57.166 ping statistics —
5 packets transmitted, 5 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 25.586/27.224/28.784/1.282 ms
anurag:~ anurag$

 

Half latency and that’s BSNL router in Delhi/Noida where they are taking drop from Tata. It’s BSNL’s router but sitting on Tata’s IP for BGP session. So this clearly tells that when we see routes from AS9829 to AS4755 Tata-VSNL they are between routers within India.

 

Now coming back to bad route between BSNL and Softlayer, in that case first few hops are:

1 [AS65534] router.home (10.10.0.1) 1.189 ms 0.910 ms 0.810 ms
2 [AS9829] 117.220.160.1 (117.220.160.1) 17.707 ms 21.147 ms 16.925 ms
3 [AS9829] 218.248.169.126 (218.248.169.126) 30.195 ms 29.766 ms 29.976 ms
4 [AS9829] 218.248.250.82 (218.248.250.82) 75.432 ms 77.488 ms 76.761 ms
5 [AS6453] if-11-1-1.mcore3.laa-losangeles.as6453.net (209.58.85.5) 368.104 ms 303.206 ms 309.964 ms

 

Hop 5 has latency of 300ms (usual for India > US routes). Again assuming 209.58.85.5 is coming from /30 and as per usual BSNL practice next IP in that subnet i.e 209.58.85.6 would be on BSNL’s side, let’s ping 209.58.85.6:

anurag:~ anurag$ ping -c 5 209.58.85.6
PING 209.58.85.6 (209.58.85.6): 56 data bytes
64 bytes from 209.58.85.6: icmp_seq=0 ttl=250 time=373.483 ms
64 bytes from 209.58.85.6: icmp_seq=1 ttl=250 time=395.493 ms
64 bytes from 209.58.85.6: icmp_seq=2 ttl=250 time=419.340 ms
64 bytes from 209.58.85.6: icmp_seq=3 ttl=250 time=305.460 ms
64 bytes from 209.58.85.6: icmp_seq=4 ttl=250 time=362.598 ms

— 209.58.85.6 ping statistics —
5 packets transmitted, 5 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 305.460/371.275/419.340/38.232 ms
anurag:~ anurag$

 

 

Hmm….300ms latency. Unexpected. I thought this router was in India but this seems slightly complex. Likely BGP session here is using BSNL’s /30 subnet and not via Tata Comm’s subnet. 

OK – let’s see last IP from BSNL on that trace – it was 218.248.250.82. Let’s ask Tata AS6453 Los Angles LAA router via AS6453 Looking Glass for BGP table:

 

Router: gin-laa-mcore3
Site: US, Los angeles, LAA
Command: show ip bgp 218.248.250.82

BGP routing table entry for 218.248.240.0/20
Bestpath Modifiers: deterministic-med
Paths: (2 available, best #1)
14 16 17 18
9829
ix-3-2.mcore3.LAA-LosAngeles. from ix-3-2.mcore3.LAA-LosAngeles. (218.248.254.99)
Origin IGP, valid, external, best
Community:
9829, (received-only)
ix-3-2.mcore3.LAA-LosAngeles. from ix-3-2.mcore3.LAA-LosAngeles. (218.248.254.99)
Origin IGP, valid, external

 

So BGP route is via – 218.248.254.99

Let’s trace:

traceroute to 218.248.254.99 (218.248.254.99), 64 hops max, 52 byte packets
1 router.home (10.10.0.1) 4.047 ms 0.875 ms 0.958 ms
2 117.220.160.1 (117.220.160.1) 18.779 ms 17.490 ms 19.334 ms
3 218.248.169.126 (218.248.169.126) 44.040 ms 32.802 ms 29.831 ms
4 218.248.250.174 (218.248.250.174) 82.626 ms 87.126 ms 84.243 ms
5 218.248.255.99 (218.248.255.99) 86.061 ms 85.503 ms 83.003 ms

 

Here we go!

So clearly BSNL on 218.248.255.99 is placed in India and is having a BGP session with Tata AS6453 router in Los Angeles. This is over an IPLC circuit of Tata Communications. 

 

Possible fix…

Following an amazing quote – “Never call it a problem unless you have the solution!

So problem here is not really via Tata’s network. They are just selling bandwidth in form of two products – IP Transit & IPLC. It’s BSNL’s wrong idea of using IPLC carelessly. Likely BSNL won’t care or put much effort in fixing it. 

There can be a possible fix from Softlayer side. If they blackhole prefix announcement to BSNL AS9829 via Tata AS6453, BSNL will never pick their IPLC (or even IP) route. Instead they will just pick route via any other upstream like Airtel or Reliance Globalcom.  

 

Let’s look at relationship of Tata AS6453 with PCCW Global (upstream for Softlayer)

anurag:~ anurag$ whois -h whois.radb.net as6453 | grep -w AS3491
import: from AS3491 action pref = 100; accept AS-CAIS
export: to AS3491 announce AS-GLOBEINTERNET
import: from AS3491 action pref = 100; accept AS-CAIS
export: to AS3491 announce AS-GLOBEINTERNET
anurag:~ anurag$

 

Clearly both are peering! 

Based on presentation from Mr Amit Dunga (from Tata Communications) at SANOG, here’s list of BGP communities used by Tata AS6453:

Screen Shot 2013-06-01 at 12.30.35 AM

 

 

Thus if Softlayer could get it’s upstream providers (like PCCW in this specific case) to use 65009:9829 – this will ensure that route learnt by Tata AS6453 from PCCW Global AS3491 is NOT exported to BSNL AS9829. Thus BSNL will instead get route via Bharti Airtel AS9498 or Reliance AS18101.

 

I just sent this detailed info as email to Softlayer and BSNL. And oh yes – I don’t know why hostgator.in is hosted in Softlayer Singapore anyways. They provide hosting in India out of Ctrls datacenter. Why they host their own home site in Singapore is something beyond my understanding!

 

With hopes that your packets to Singapore are not routing via US, time for me to get back to my “cramming” for exams. 🙂