0

Ultralow/anycast routing picks Mumbai (bom) over much closer Delhi (del) PoPs

I ran the ping test at ping.nextdns.io from Chandigarh, India, and the results clearly show Delhi PoPs beating Mumbai and Bangalore by a wide margin:

anexia-del (IPv6): 10 ms

vultr-del: 13 ms

vultr-del (IPv6): 13 ms

anexia-del: 14 ms

vultr-bom (ultralow2): 35 ms

ls-bom (ultralow1): 35 ms

vultr-bom (IPv6, ultralow2): 36 ms

ls-bom (IPv6, ultralow1): 37 ms

anexia-maa (anycast1): 48 ms

vultr-blr: 48 ms

do-blr: 52 ms

Despite this, my actual DNS traffic is consistently routed to ls-bom/vultr-bom (the ultralow1/ultralow2 PoPs), not the Delhi PoPs, which are 20-25 ms faster and geographically much closer to me. This has been consistent across multiple checks, not a one-off blip.

I've run the diagnostic tool and here's the report: https://nextdns.io/diag/9df0f8d0-a35c-11f1-b858-e31f6ee4f0bd

Could someone take a look at why ultralow routing is preferring Mumbai over Delhi for my network/ASN? Happy to provide additional traceroutes (to 45.90.28.0/45.90.30.0, 2a07:a8c0::/2a07:a8c1::, and ipv4.dns1/dns2.nextdns.io) if useful.

1 reply

null
    • AYM
    • 13 hrs ago
    • Reported - view

    Update, ~14 hrs later: Re-ran everything — same core issue, plus one more data point worth adding.

    ping.nextdns.io now explicitly tags vultr-del/anexia-del as (ultralow1)/(ultralow2) — so your system has itself identified Delhi as the correct ULL PoPs for my network. Yet ULL Primary/Secondary steering still resolves to ls-bom/vultr-bom (ultralow on Mac diag, no ultralow tag on ping.nextdns.io). Manually querying ultralow.dns1.nextdns.io and ultralow.dns2.nextdns.io directly still returns bom, not del — so this isn't client-side caching, it's the resolver itself.

    Anycast is worse: Secondary (v4) resolves to zepto-tpe in Taipei at 127-204 ms (via Hong Kong through Global Secure Layer), and Secondary (v6) resolves to vultr-ams in Amsterdam at ~154-156 ms. Both are massive detours for an Indian query when Delhi/Mumbai PoPs exist and measure under 60 ms on the same network.

    Given ULL, anycast v4, and anycast v6 are all mis-steering for the same ASN (AS9829, BSNL, Chandigarh), this looks less like an isolated glitch and more like the whole India-region routing/GeoIP mapping needing a review.

    Fresh diag: https://nextdns.io/diag/b2a04080-a3b9-11f1-a2ce-d15b343e9471

    Could someone take a broader look at routing for this ASN/region?

Login to reply

Content aside

  • 2 hrs agoLast active
  • 1Replies
  • 19Views
  • 1 Following