yeah, using Airtel only.
@Anurag Bhatia I have a weird issue with Airtel on Ookla, can I discuss with you?
@Anurag Bhatia I have a weird issue with Airtel on Ookla, can I discuss with you?
-
Amazon Deals - ToS - Like us?
Excitel Fiber Broadband Experience in Delhi NCR
- Thread starter Life_Warrior
- Start date
- Replies 407
- Views 48,105
@Anurag Bhatia I have mailed you, please check.
Haha. Not that I agree to that conclusion but thanks for the kind words. I see the post to which you replied this is deleted now.
Anyways, some updates on more testing. It's 5am right now. Absolutely off peak time. I wanted to test more around this time to find possible policies impacting speeds without getting impacted due to possible congestion.
I ran a dozen of iperf3 tests against three different servers of mine. On my 400Mbps plan, this is what I get (testing via my Linux desktop connected over wire ofcourse):
Single thread:
Twenty threads:
This is likely due to one of two reasons
- There is a (DPI?) device shaping traffic flows or some policy on devices facing upstream shaping the traffic
- They have too many LACP bundles of 10G ports to aggregate bandwidth and some of these are choking. Multiple threads go over different bundles reducing chances of having only one stream over choked port.
If #2 is true, we should not see this behaviour for outbound traffic considering fact that Excitel is heavily an inbound network. Their outbound direction would be lying almost empty (20-30% of what their inbound flows would be like). So test again but for uploads this time:
Single thread uploads:
Twenty thread uploads
Again similar behaviour. But since ICMP stays very stable even to far off locations, this hints use of some sort of overall traffic shaping policy.
@Anurag Bhatia Legend himself on this forum. Hello sir, Good to see you here 
.
.
Haha. Nothing like that. Thanks for your kind words and greeting @unnecessary@Anurag Bhatia Legend himself on this forum. Hello sir, Good to see you here
That seems overall OK for the specs. Power can go as low as -27 or even -28dbm for most cases without causing issues. However they (LCOs in this case) try for much better margin numbers of -22 to -23. This gives sufficient margin for issues in cables (common due to monkeys & what not for overhead cables) and also give safer margin to split near subscriber end as well as in the backend for newer connections. Except Jio & Airtel most of other networks add PON splitters on the go. They won't simply do a 8 or 16 split per pole. For Jio that made sense based on the scale, anticipation of users and an overall design which works across country (without having to re-do it again and again).
For smaller players that would be quite a bit of cost to do that in advance. On their advantage - smaller players can react faster than large company like Jio. For smaller networks, LCO based models - you can expect higher power as a smaller network is deployed and it will reduce over time as they add more splits to add customers.
Correction on my previous conclusion on expectation on peered routes
Earlier I mentioned:
As I used Excitel more, I have come to conclusion that even peered routes are having traffic shaping issues and do not give expected/sufficient bandwidth. Here's a demo of push of 4GB dump to AWS S3 in Mumbai which goes over Excitel-AWS peering in Delhi.
Single uploads peak at ~72Mbps
I see similar speeds even when doing 20 parallel transfers.
And again to be sure this isn't some NLD congestion between Excitel's PoP at STT / Tata Comm DC in Delhi and Rohtak, I see 400Mbps symmetric on both speedtest.net and fast.com or any other issue say on my home LAN.
This differs significantly than my primary connection from a local ISP (IAXN). This is a 150Mbps connection (Ookla test here). Upload performance on 150Mbps plan:
IAXN 150Mbps plan, similar upload to AWS S3 at ~ 136Mbps
Could AWS S3 Mumbai could be bottleneck here?
Here's test from my dedicated server in Equinix Mumbai. Server connects to same router which has a PNI with AWS in Mumbai and server is 0.2ms from AWS S3 endpoint:
This confirms max possible speed of atleast 664Mbps here way above 400Mbps Excitel plan which I am testing.
Conclusion
Thus sadly while Excitel seems one of best networks latency wise in my area (due to choice of their NLD), it's barely useful for anything heavy. I save 4-5ms on latency but loose out completely on the performance.
Excitel has probably priced plan low enough for now that they are making it work by treating popular speedtests (Ookla, fast.com etc) differently than actual traffic. Made it my backup provider for next three months, will re-test a week before end of three months and likely will just drop it (unless something dramatic changes).
