JioFiber Mumbai Performance Review

[Lolita_Magnum]

Temporarily had to move to a friend's place due to Covid, had been using JioFiber here.
Here's the overview.
Plan 150Mbps + OTT and STB.
Speeds are always 90% of the mentioned speed or lower.
Pings are acceptable but not liking high hops, and compulsory routing via Singapore for overseas.
Although downloads from overseas servers are quite stable.

(unacceptable loaded latency)

Verdict: won't recommend if you do latency critical work, or have application that rely on latency while being on load.
Uptime: 99.9% of the time, so for Avg user I guess it's good.

 

[Anurag Bhatia]

Pings are acceptable but not liking high hops

Hop count doesn't matter anymore due to MPLS. You can find a network with less hops and high latency & vice versa because MPLS tunnels often hide the middle network on layer 3 based traceroute.


Take e.g trace from California to Haryana

traceroute to 202.9.120.253 (202.9.120.253), 30 hops max, 60 byte packets
1 _gateway (10.50.60.1) 0.390 ms 0.316 ms 0.230 ms
2 10.50.50.1 (10.50.50.1) 0.440 ms 0.330 ms 0.374 ms
3 v1388.core3.fmt1.he.net (65.19.151.9) 0.688 ms 0.608 ms 0.642 ms
4 port-channel2.core2.pao1.he.net (72.52.92.65) 1.516 ms 1.481 ms 1.634 ms
5 100ge1-2.core1.pao1.he.net (184.104.195.173) 1.236 ms 1.253 ms 1.202 ms
6 100ge5-2.core1.lax2.he.net (184.105.81.102) 9.748 ms 9.950 ms 9.851 ms
7 bharti-airtel.as9498.any2ix.coresite.com (206.72.210.149) 10.122 ms 10.120 ms 9.956 ms
8 182.79.152.120 (182.79.152.120) 259.451 ms 182.79.181.93 (182.79.181.93) 436.011 ms 182.79.152.0 (182.79.152.0) 270.060 ms
9 182.79.198.202 (182.79.198.202) 257.311 ms 182.79.198.206 (182.79.198.206) 263.205 ms 182.79.198.214 (182.79.198.214) 431.618 ms
10 182.74.20.10 (182.74.20.10) 282.638 ms 283.252 ms 282.467 ms
11 axntech-dynamic-253.120.9.202.axntechnologies.in (202.9.120.253) 265.831 ms 261.448 ms 257.760 ms

Here hop 7 is Airtel in Los Angeles and hop 8 is Airtel in Delhi NCR. So just 1 hop and we crossed Pacific Ocean + Asian network + even Indian network from Chennai to Delhi NCR. Traffic here ofcourse is via a bunch of routers in middle. They are just not visible because of the MPLS tunnel. One can still run a network without MPLS tunnels and many hops would be visible without any impact on IP performance here.

compulsory routing via Singapore for overseas.

What makes you think so? Please do keep in mind that physically India connects to Singapore & Middle East via submarine cables.

The world’s largest network and datacenter infrastructure atlas

www.infrapedia.com

So for Eastern traffic towards Singapore, Japan, US West and even rest of Asia Singapore is a key location. Plus market is quite developed with great infrastructure, good datacenters and lot's of interconnection. Thus it is very much expected to see Singapore for traffic in Eastern direction.

For traffic towards West typically you will find traffic routing via Europe and not specifically countries in Middle East even though submarine cable path has those. It has to do with peering ecosystem, overall flow of traffic demand etc.

 

[ausyvthb]

Can you do a packet loss check from packetlosstest.com ? I am curious about that.

 

[Lolita_Magnum]

Hop count doesn't matter anymore due to MPLS. You can find a network with less hops and high latency & vice versa because MPLS tunnels often hide the middle network on layer 3 based traceroute.


Take e.g trace from California to Haryana


Here hop 7 is Airtel in Los Angeles and hop 8 is Airtel in Delhi NCR. So just 1 hop and we crossed Pacific Ocean + Asian network + even Indian network from Chennai to Delhi NCR. Traffic here ofcourse is via a bunch of routers in middle. They are just not visible because of the MPLS tunnel. One can still run a network without MPLS tunnels and many hops would be visible without any impact on IP performance here.





What makes you think so? Please do keep in mind that physically India connects to Singapore & Middle East via submarine cables.

The world’s largest network and datacenter infrastructure atlas

www.infrapedia.com

So for Eastern traffic towards Singapore, Japan, US West and even rest of Asia Singapore is a key location. Plus market is quite developed with great infrastructure, good datacenters and lot's of interconnection. Thus it is very much expected to see Singapore for traffic in Eastern direction.

For traffic towards West typically you will find traffic routing via Europe and not specifically countries in Middle East even though submarine cable path has those. It has to do with peering ecosystem, overall flow of traffic demand etc.

Good for you, even though mpls doesn't affect ip performance, with jio have extra natting over each does hit performance, even loaded latency at times are considerably high.
This is not a issue, with lot other isps.
And, I'm saying jio already has lot of hops, which does detoriate the network, you are not using, but I am, and I see massive ping spikes at times and unstable connections, secondly there are times when it takes atleast 60ms, just for it leave the internal class c network to reach 1.1.1.1, while other isps here can do it in just 2ms.
And for the compulsory jio route via Singapore, for example, I use leaseweb for my server needs, and when I traceroute to my server, it always routes via Singapore and has an added latency 320ms, while for my other isp, that routes via sometimes Tata or airtel, gets it done in just 139ms! That's a considerable difference to not use jio.
For the packet loss test, it's 0% packet loss.

 

[Anurag Bhatia]

At risk of repeating again - hops do not add latency. Your issue with Leaseweb and others is just bad routing case. You can have bad routing with or without extra hops and MPLS. I understand your frustration with latency and yes I am not using Jio FTTH or claiming it's good. But just correcting your assumption about the issue.


Regarding Leaseweb and similar cases, these can happen in dense mesh peering without optimisation.

I looked at trace to leaseweb01.ring.nlnog.net (Haarlem, NL) and it looks good from active RIPE Atlas probes on Jio. That's 150ms or less.
Also for leaseweb04.ring.nlnog.net (Frankfurt, DE) also seems mostly okish.

Where is your server located? Can you share IP or subnet for trace?

 

[Lolita_Magnum]

Leaseweb Netherlands 212.7.207.214.
Alright, how do you explain this?

(look at that bs, until hop 15)

 

---

[JB701]

@Lolita_Magnum Can you do 85.17.182.245 ?

 

[Anurag Bhatia]

Leaseweb Netherlands 212.7.207.214.
Alright, how do you explain this?

(look at that bs, until hop 15)

Whenever latency jumps suddenly at certain hop then that's often because of bad return path. In this specific case Cloudflare likely not having a return path within India leading to it being routed from outside of India. It can be due to number of reasons and need to explored further along the lines:

1. If Jio announcing prefix covering your CGNAT WAN IP to Cloudflare?
2. If yes to #1 then is Cloudflare accepting it? If not, why not?
3. This specific PoP might be inside Airtel and in that case Cloudflare is expected to return traffic here via Airtel. If Airtel is missing direct domestic route for this covering prefix - again it will go outside of India.


To get this fixed share the WAN IP. WAN IP pool is fine. You can visit Hurricane Electric BGP Toolkit and share output of the pool. I will request folks at Cloudflare to share the return path. 99.9% sure this is bad return route and not exactly a congestion issue.