What is the issue you’re encountering
For the past two weeks, there has been high latency on Airtel for internet traffic to and from Europe. Airtel, a major ISP in India, is used by Cloudflare as the main upstream provider for Warp, with some routes leveraging Cloudflare’s own backbone.
This issue of high latency has affected all European and some US IP addresses when using Zero Trust Warp. Over the past two weeks or more, ISPs and individuals relying on Airtel in India as their upstream provider have experienced significant latency to Europe. Reverse traceroutes from Germany to Indian endpoints revealed that Airtel was rerouting traffic via Singapore and the USA, leading to increased latency. However, the latency spike appeared understated in charts because the data included endpoints from unaffected ISPs such as Tata Communications, Jio, Sify, and Vodafone IDEA. Airtel has not provided a explanation for this routing behavior.
You can refer to the “airtel” image I have shared, taken from Airtel routing Europe to India via the US - Personal blog of Anurag Bhatia
On December 23rd, latency returned to normal levels of around 130–140 ms (see the “airtel_1” image), depending on the endpoint. This normalization applied to ISPs using Airtel as their upstream, data centers, and Airtel broadband itself. However, Cloudflare Warp, which heavily relies on Airtel, still faces abnormal latency.
The high latency on Warp is caused by traffic being routed via Singapore or the USA to Cloudflare’s Mumbai PoP instead of taking the direct path, as it did two weeks ago when everything was functioning normally.
To clarify, this issue arises while using Zero Trust Warp, and not due to being connected to a distant data center. I am connected to the closest data center, BOM (Mumbai, India). Initially, the issue seemed limited to Airtel. However, incoming traffic on TATA (AS6453) is also being routed via Singapore to Mumbai on TATA before reaching Cloudflare’s Warp Mumbai colocation. This is peculiar, as public traceroutes(run on many probes) for TATA showed normal routing even during Airtel’s high latency period.
This morning, I observed that traffic routed via Cloudflare’s backbone to destination IPs in Europe and the USA is now going via Singapore(take a look at the traceroute below) instead of directly from Mumbai PoP to Europe, which was the case earlier. Currently, all traffic appears to be routed through Singapore to Europe and the USA when using Cloudflare Zero Trust Warp.
xxx@xxx-xxx-xxx ~ % sudo mtr -wzrb 116.202.235.227
Start: 2024-12-27T20:15:28+0530
HOST: xxxxx-xxxxx-xxxxx.local Loss% Snt Last Avg Best Wrst StDev
1. AS13335 172.71.201.68 0.0% 10 7.8 11.0 7.8 13.5 1.9
2. AS13335 104.28.0.0 0.0% 10 11.9 12.0 8.3 13.6 1.5
3. AS??? ??? 100.0 10 0.0 0.0 0.0 0.0 0.0
4. AS13335 162.158.226.76 0.0% 10 13.2 15.8 9.9 34.7 6.9
5. AS??? 198.41.160.16 90.0% 10 76.3 76.3 76.3 76.3 0.0
6. AS??? 198.41.140.243 40.0% 10 316.6 315.9 311.7 323.8 4.4
7. AS??? mrs.de-cix.gw.hetzner.com (185.1.47.15) 10.0% 10 204.1 204.0 200.0 205.5 1.6
8. AS24940 core21.fsn1.hetzner.com (213.239.224.105) 20.0% 10 206.2 206.8 205.6 207.6 0.7
9. AS24940 ex9k1.dc15.fsn1.hetzner.com (213.239.252.114) 10.0% 10 209.0 207.9 204.4 209.0 1.4
10. AS24940 static.227.235.202.116.clients.your-server.de (116.202.235.227) 20.0% 10 210.3 209.7 207.5 210.3 1.0
Is there any ETA on when traffic between Mumbai and Europe will be routed normally? As far as I know, Airtel and TATA do not have ongoing issues, and on Cloudflare backbone, traffic continues to be routed through Singapore to Europe and the US.
