Should we call these transit providers?
Yes, that would be correct term.
Also, why did Alliance announce my IP thrice?
That is usual traffic engineering. More specific /24 is always preferred by the networks, less specific /23 is less preferred and least specific /22 is least preferred. This way an ISP can announce most specific at exchanges/peering, less specific say /23 to their preferred transit provider (say
Tata Comm in this case) and least specific to least preferred transit provider (Airtel in this specific case). Thus if any issue happens on their Tata Comm circuit, world will just start sending them traffic via Airtel instead of a full blackout for those IP pools.
If a caching node/ direct peering was broken, why didn't they redirect the traffic to one IXP/ transit instead? Wouldn't that solve the issue?
That happens by default. If a working caching node gets any issue, traffic will just go to nearby IX point. If that IX link is broken, it will go via their IP transit provider. As a matter of fact many of these large players like
Google, FB are too picky on packet loss and would automatically re-route traffic via different path if they see congestion. ISPs typically control things based on prefix announcement via BGP but CDNs often add
DNS layer to it. Thus something.something.static.youtube.com may resolve to Alliance local GGC cache for you and if there's an issue, next time it may start resolving to GGC cache node inside Airtel in Kolkata or Mumbai. You may want to read
this post of mine which briefly covers that in Google's context and
this talk by Netflix VP covers that for
Netflix.
Can latency jump from 6 to 12 to 18 ms due to congestion?
Yes. It can happen. Usually during congestion you will see latency changing lot more often. So it would be 6,10,80,20 etc. Instead of 6-7-6-8 etc. That's because large links typically do not congest by just one traffic flow. Its the flow of traffic from hundreds/thousands of subscribers. Thus if an ISP has real world traffic of say 1.1Gbps & they are pushing it over say a 1Gbps link, you won't see flat line on the traffic all the time. That's because 1.1Gbps will be like 1.1Gbps at 0th second, 1.01Gbps at 1st second, 900Mbps at 2nd second, 950Mbps at 3rd second etc. You will typically see consistently high latency & a flat graph at ISP end if say one is trying to fit 2-3Gbps over a Gig link where that specific link stays at 100% usage all the time.
why is there congestion to Delhi/ Mumbai and that too, specifically for my town?
It's hard for me to comment on that without having a view of overall topology, capacity etc. Are you still having issues presently? If yes then drop me a mail on
me@anuragbhatia.com with your WAN IP, couple of speedtest, couple of traces towards those end points and I will connect you with friends in Alliance who might be able to help. If issue is solved then all I can say is that it might be congestion or unexpected bad routing etc. In some cases I have seen network operators missing to announce all customer pools on peering and thus unexpected traffic hits their transits and causes congestion. Your neighbour under same LCO and even same
GPON branch can have different routing beyond the ISP gateway depending on the WAN IP. If both of you get WAN IP from same pool then routing would be similar or if both are NATed behind same IP/pool then again routing would be similar. But if WAN IPs are unrelated/different things can take totally different path.