-
Amazon Deals - ToS - Warp
IS there any way to detect number of users sharing the same PON cable?
- Thread starter Realme
- Start date
- Replies 13
- Views 1,477
LCO can only see/detect number of active users that too only in OLT UI Other than that there is no known method for detecting number of active users.
So if 10 users take 300Mbps and all burst at the same time - remaining 54 users will suffer.
Firstly...
Though this is highly non-practical condition. I know it may seem odd to be saying this on a forum of tech enthusiasts but average users do not use that much bandwidth. They may have 300Mbps plan but besides doing speedtest which will burst the circuit for few seconds, most of average users will really not fill the up the pipe. Even when you add 2-3 4K TVs (25Mbps stream each). Networks are typically designed keeping average as well as long peak bursts in mind and not super short peak bursts. Same logic applies even to say 2 billion WhatsApp users or a billion plus of YouTube users. They won't design the capacity looking at just the law number of users but realistic use of bandwidth by these users.
Coming to your question - assuming for whatever use case, 10 people actually manage to saturate the GPON branch, in that case ISP can always "lit" 2-3 more strands from the OLT till the junction box nearby and move some users on different GPON port and that would take care of the congestion. That's short to mid term solution. In long term, 10G PON will become common and each branch would have a mix of 2.5G and 10G running on different wavelengths until everyone is moved to 10G PON (few years from now).
Firstly...
Though this is highly non-practical condition. I know it may seem odd to be saying this on a forum of tech enthusiasts but average users do not use that much bandwidth. They may have 300Mbps plan but besides doing speedtest which will burst the circuit for few seconds, most of average users will really not fill the up the pipe. Even when you add 2-3 4K TVs (25Mbps stream each). Networks are typically designed keeping average as well as long peak bursts in mind and not super short peak bursts. Same logic applies even to say 2 billion WhatsApp users or a billion plus of YouTube users. They won't design the capacity looking at just the law number of users but realistic use of bandwidth by these users.
Coming to your question - assuming for whatever use case, 10 people actually manage to saturate the GPON branch, in that case ISP can always "lit" 2-3 more strands from the OLT till the junction box nearby and move some users on different GPON port and that would take care of the congestion. That's short to mid term solution. In long term, 10G PON will become common and each branch would have a mix of 2.5G and 10G running on different wavelengths until everyone is moved to 10G PON (few years from now).
Backbone wise running 10G/40G/100G to cities is quite feasible. Bandwidth is typically not major challenge on backbone level as long as one can pull sufficient fiber in. It's usually a challenge where fiber is not present or hard terrain hilly areas or when it's for last mile in low density areas.
With DWDM one can easily increase bandwidth to great extent.
This applies to BSNL, Airtel, Jio or just any major operator. It's purely case of demand and supply. ISPs could do 8Mbps easily on most of old DSL infra but they did not do for a long time and instead had plans from 256Kbps to 2-4Mbps. In most of cases it's the economics behind the plan. How many extra users do you think BSNL can get if they offer 1Gbps for an attractive pricing of say 700-800Rs a month? How many of those users would only purchase service if they can get 1Gbps and will not consider 50-100-200Mbps kind of plans? For now answer to those questions is - "Not that many". If you check their plans, they do offer 300Mbps for 1499 a month. If they offer 1Gbps for 700Rs, what happens to this premium plan? Should they give 3Gbps on that? That's hard one because home interfaces, PON CPE side interface etc is all 1Gbps. Plus there's a high chance person would just downgrade from 1499 to 700Rs plan and live happily with 1Gbps. So at the core of it you have to ask question "What is one an do with 1Gbps which one cannot do with 100Mbps". Right now not much but as that changes, you can expect 1Gbps plan. In absence of that offering 1Gbps would just put a bill of upgrades, extra capacity to serve peak with no realistic returns.
This has been discussed extensively on this forum. Try searching for previous discussion or you can check my blog post about their routing issues when I left their service - BSNL AS9829 – A rotten IP backbone
Broadly it's due to following:
- BSNL is quite poorly peered compared to other networks inside India. As a large eyeball and inbound heavy networks, they should have been peered with all large content players but that's the case.
- BSNL has IPLC - basically long point to point circuits from India to outside world and they took large capacity on circuits to New York and Los Angeles. They take different upstream providers in those locations and thus end up with many circuits, many providers. To fill up on this capacity they announce specific prefixes and that results in cases where for any network in the world (outside of India) to send traffic to BSNL, only entry point happens to be a gateway in New York or Los Angeles. This impacts routing from nearby areas including Singapore, Hong Kong and even Europe.
- While other large networks of that size typically peer outside of India, BSNL never peers outside of India and just buys capacity. Peering by design keeps traffic local and typically would keep routing clean for 99% part.
- BSNL doesn't uses BGP communities on their circuits outside of India and use one set of provider at one location & other set at a different location.
- Airtel doesn't accept BSNL routes in South India and in many of those instances routes to BSNL from outside of India. That again results in issues of high latency.
Thus it's lack of overall traffic engineering, good practice etc which creates a not-so-good routing.
@Anurag Bhatia
I have read your blog earlier and checks it out for new posts.
I had few questions regarding the working of BSNL.
1. If they have bought bandwidth, it is definitely possible for them to peer outside of India with various networks and cloud services having straight control ove routing. What exactly is the resistance point because AFAIK it's just a little configuration to be completed on any of their end router which announces routes? Moreover, the termination of most circuits is an exchange (my assumption).
2. What exactly is the issue with buying circuits? I assume they still have control over how they route their traffic based of the target IP location. So, It should still work and work even better since they basically have dedicated capacity from one end to another. Is it just stupid personnel who are lazying around instead of creating better routing rules or this "buying bandwidth" has some effect on it?
I have read your blog earlier and checks it out for new posts.
I had few questions regarding the working of BSNL.
1. If they have bought bandwidth, it is definitely possible for them to peer outside of India with various networks and cloud services having straight control ove routing. What exactly is the resistance point because AFAIK it's just a little configuration to be completed on any of their end router which announces routes? Moreover, the termination of most circuits is an exchange (my assumption).
2. What exactly is the issue with buying circuits? I assume they still have control over how they route their traffic based of the target IP location. So, It should still work and work even better since they basically have dedicated capacity from one end to another. Is it just stupid personnel who are lazying around instead of creating better routing rules or this "buying bandwidth" has some effect on it?
@Realme
AFAIK, there are equipments available in market specifically able to listen to the optical signals. They usually send a packet "which dissociate the connection" forcing a reconnection request and catches the encryption key, essentially listening to most of the traffic. But they are useless until you have access to root trust provider, which can give you access to keys.(I still suspect that there are back doors in most hardwares😅)
You can even re-configure an ONT to listen to all the traffic or to keep sending packets to block services of other users. You would just need access to documentation of the hardware and some knowledge of the programming.(I know javascript😁, I know it's not enough)
AFAIK, there are equipments available in market specifically able to listen to the optical signals. They usually send a packet "which dissociate the connection" forcing a reconnection request and catches the encryption key, essentially listening to most of the traffic. But they are useless until you have access to root trust provider, which can give you access to keys.(I still suspect that there are back doors in most hardwares😅)
You can even re-configure an ONT to listen to all the traffic or to keep sending packets to block services of other users. You would just need access to documentation of the hardware and some knowledge of the programming.(I know javascript😁, I know it's not enough)
This would need a long answer thus excuse a longer reply.
1. Peering outside India is surely doable but practically it's not that straightforward for a Govt. operator. Do not under estimate logistics as well as Govt. policies around purchase of colocation, hardware, services etc. BSNL has circuits from India to Los Angeles and they just buy transit there. If today they decide to go ahead with peering at say any large IX over there, say Equinix Los Angeles then they have to do a few things: Buy colocation. Can they legally just pick and decide a datacenter colo, I doubt that. Most of things would require to call for a tender/invitation for bidding. Will Equinix and other datacenters will pick that text heavy Indian paper work, may be yes, may be no. So you see it's hard for the decision maker to justify specific datacenter as well as specific IX. I think they have some workaround but it remains hard bureaucracy wise. Assume that part is navigated, next you hit issue: American datacenter would typically expect money in USD and not INR to avoid currency exposure. They would typically not sell/do contract in INR because if INR losses value over short periods, it can cause issues for them. Can BSNL negotiate in USD? Would be hard as would need approval of dept. outside of BSNL like finance ministry and even RBI to issue payments. Assume this part is navigated, next they have to deploy hardware - essentially a router & atleast a switch etc. How you do that? Can you ship hardware from India? That would be expensive and even counterproductive. Can you buy or lease locally - well again how you decide vendor? At each step where purchase is required they will not have the free hand. Assume this part is navigated. Next comes - how do you install and maintain that gear which is far away from India? Again you need local vendor/RHS support etc. So do not underestimate the fact that lot of logistics is involved here. Plus BSNL would typically not have free hand in making purchases as the private players. Lack of free hand introduces inefficiency and giving a fully free hand may likely lead to corruption. So it's a hard problem to solve.
2. It's not an issue if BSNL peers with networks at those locations instead of buying IP transit/wholesale bandwidth. Peering is hard as I explained in above due to fact that they are a Govt. organisation. If they pick IP transit outside of India BGP will not always pick best path. Have a look at slide 27 here to read BGP path selection algorithm. So looking at routing right now I see BSNL has IP transit from Airtel, Tata Comm (domestic), Tata Comm (outside of India), Sri Lanka Telecom, Cogent, GTT, Telia, Century Link etc. Let's pick any one specific transit and find any specific prefix announced there say from GTT AS3257.
(and more)
GTT is American player & from looking at routes, I can confirm BSNL is taking this transit in the US. Let's now trace from a few locations in the Europe. Ideally it should be Europe > India as direct submarine cables exist. Let's read a few traces:
DigitalOcean UK > BSNL India
Custodian datacenter AS50300 > BSNL India
Let's pick somewhere else in Europe. Say Germany & Netherlands
AS62874 Germany > BSNL India
AS12859 Netherlands > BSNL India
So it's just a endless list. In all these cases traffic is going from the respective country in Europe (from where we initiate trace) to BSNL India via the US. ae3.cr6-nyc6.ip4.gtt.net - tells that it's New York router of GTT and hence most of traffic here is routing via New York. So instead of latency being 130-170ms (depending on where the destination is in India) it's as high as 250ms.
Why BGP is taking this unoptimised path. That's due to the fact that BSNL is announcing routes only in the US via these large networks and not in Europe. Hence for anyone to send traffic entry point is only the US. Besides GTT, for the same pool BSNL is also using Tata Comm AS6453 but again outside of India. Here are a couple of traces to show the same:
(multi part reply to stay in word limit)
(rest of reply)
Leaseweb Germany AS28753 > BSNL India
Linode London AS63949 > BSNL India
Besides only transit there, another factor BGP would very often use is preference of downstream customers over peers (see slide 29 here). So even if say BSNL taking IP transit from Tata Comm AS4755 in India (which they do have actually) and announce routes to that as well as over this GTT circuit in New York, then entire GTT network as well as their single homed customer (who only buy from GTT) would always prefer GTT > BSNL path no matter even if has to go a few continents. That AS_PATH is shorter, plus upstream here (GTT) would have high localpref on routes via a customer (BSNL) where they make money compared to same route from a peer like Tata Comm which very likely is settlement free peer and where they won't make money by sending traffic. BSNL can in theory tweak it and control a bit i.e by using BGP communities. They could signal to transits in the US (GTT/Tata Comm in this example) to announce that route only to their peers and customers in the US, South America etc and not in Europe, Singapore etc and have setup in place to take care of traffic from Europe.
With all that being said - please also understand that while there are 65000+ ASNs/networks the world, typically top 20-30 carry as much as 90% of traffic towards an eyeball like BSNL, Jio, Airtel, MTNL, ACT etc via peering as well as caching nodes. And out of those top 20-30, BSNL can easily peer with top 15-20 within India. Once you are peered to those for remaining 64970 ASNs you are looking at 10-20% of traffic and that too diminishing as more traffic aggregates to large CDNs and cloud player (for good or bad, whatever but that's happening). Thus I would say BSNL could just do good domestic peering, drop all long haul ILD circuits and take local IP transits to save from management as well as traffic engineering efforts. That's more realistic, easy and efficient. There's a funny ratio in our industry - bandwidth cost is like 1:3:10. So if at core it's $1, it would $3 on middle mile and $10 at last mile (where end user is connected). Thus whether BSNL keeps inefficient routing while keeping it's purchase price at $3 or $2.8 that matters less if users paying $10 to them start leaving due to latency issues.
Leaseweb Germany AS28753 > BSNL India
Linode London AS63949 > BSNL India
Besides only transit there, another factor BGP would very often use is preference of downstream customers over peers (see slide 29 here). So even if say BSNL taking IP transit from Tata Comm AS4755 in India (which they do have actually) and announce routes to that as well as over this GTT circuit in New York, then entire GTT network as well as their single homed customer (who only buy from GTT) would always prefer GTT > BSNL path no matter even if has to go a few continents. That AS_PATH is shorter, plus upstream here (GTT) would have high localpref on routes via a customer (BSNL) where they make money compared to same route from a peer like Tata Comm which very likely is settlement free peer and where they won't make money by sending traffic. BSNL can in theory tweak it and control a bit i.e by using BGP communities. They could signal to transits in the US (GTT/Tata Comm in this example) to announce that route only to their peers and customers in the US, South America etc and not in Europe, Singapore etc and have setup in place to take care of traffic from Europe.
With all that being said - please also understand that while there are 65000+ ASNs/networks the world, typically top 20-30 carry as much as 90% of traffic towards an eyeball like BSNL, Jio, Airtel, MTNL, ACT etc via peering as well as caching nodes. And out of those top 20-30, BSNL can easily peer with top 15-20 within India. Once you are peered to those for remaining 64970 ASNs you are looking at 10-20% of traffic and that too diminishing as more traffic aggregates to large CDNs and cloud player (for good or bad, whatever but that's happening). Thus I would say BSNL could just do good domestic peering, drop all long haul ILD circuits and take local IP transits to save from management as well as traffic engineering efforts. That's more realistic, easy and efficient. There's a funny ratio in our industry - bandwidth cost is like 1:3:10. So if at core it's $1, it would $3 on middle mile and $10 at last mile (where end user is connected). Thus whether BSNL keeps inefficient routing while keeping it's purchase price at $3 or $2.8 that matters less if users paying $10 to them start leaving due to latency issues.
Similar threads
