About 45-ish bytes per 1500 bytes packet (I don't recall the exact numbers, I only got the 1454 bytes per packet figure in my mind). Maybe that included other overhead caused by ATM, it's been years since I've looked into this I'm afraid.
From my experience with DSL 3-5% experienced loss would make sense. I actually thought that the real experienced loss in throughput could have been up to 10% or more depending on the equipment and how the WAN to LAN routing is set up. There is definitely a noticeable impact because I have had DSL with and without PPPoE and there is a difference. I can't imagine it being much less than 3% nor have I heard of it being an insignificant percentage of the total but who knows? If it wasn't a factor then its amazing that everyone in the DSL and cable forums that rate the ISPs have talked about it for more than a decade as being a factor. Maybe it is much less...more around .5% as rotor noted but that doesn't explain why it's so noticeable of a hit.
They didn't used to do this but with all the competition now from cable it seems that many of the ISPs who are providing DSL with PPPoE now days might actually be capping their speed tiers a notch above their advertised throughput to compensate for those various losses... You can usually read long threads in the DSL and cable forums that rate ISP services about these issues regarding packet loss and the degradation in throughput for DSL services with PPPoE. I wonder if the majority of the throughput losses that people have experienced in the past have always been partially due to the extra processing required.
From the above thread link about PPPoE DSL Overhead Questions...."We know that PPPoE overhead takes about 15% off of the speed profile that a customer is on a customer with a 5 Meg plan will see around 4.25 Meg/s during a speed test."... and what Merlin said about ATM is mentioned... "BTW its 9.4% of ATM, 1.65% of TCP, 1.3% of IP, and .53% of PPPoE of overhead giving a total of 12.9%"
Last edited by SoCalReviews; 01-28-2013 at 07:48 AM.
Reason: I found more information relevant to the discussion
From my experience with DSL 3-5% experienced loss would make sense. I actually thought that the real experienced loss in throughput could have been up to 10% or more depending on the equipment and how the WAN to LAN routing is set up. There is definitely a noticeable impact because I have had DSL with and without PPPoE and there is a difference. I can't imagine it being much less than 3% nor have I heard of it being an insignificant percentage of the total but who knows? If it wasn't a factor then its amazing that everyone in the DSL and cable forums that rate the ISPs have talked about it for more than a decade as being a factor. Maybe it is much less...more around .5% as rotor noted but that doesn't explain why it's so noticeable of a hit.
They didn't used to do this but with all the competition now from cable it seems that many of the ISPs who are providing DSL with PPPoE now days might actually be capping their speed tiers a notch above their advertised throughput to compensate for those various losses... You can usually read long threads in the DSL and cable forums that rate ISP services about these issues regarding packet loss and the degradation in throughput for DSL services with PPPoE. I wonder if the majority of the throughput losses that people have experienced in the past have always been partially due to the extra processing required.
From the above thread link about PPPoE DSL Overhead Questions...."We know that PPPoE overhead takes about 15% off of the speed profile that a customer is on a customer with a 5 Meg plan will see around 4.25 Meg/s during a speed test."... and what Merlin said about ATM is mentioned... "BTW its 9.4% of ATM, 1.65% of TCP, 1.3% of IP, and .53% of PPPoE of overhead giving a total of 12.9%"
That discussion thread is full of inaccuracies and incorrect terminology and comparing apples with oranges (the opening sentence of the original post "we know that PPPoE overhead..." is completely wrong at multiple levels).
For example, the overhead of AAL5 (which sits on top of ATM), varies between 8 bytes and 55 bytes per IP packet transmitted, and the size of an IP packet is variable, so right there you see a slight problem with working out a percentage overhead, when both the overhead and the payload sizes vary as the payload has to be fit into 48-byte ATM frames.
Strictly speaking, the discussion was about the overheads of PPP, and PPPoA imposes a 2 byte overhead per IP packet, that is a fact. 2 / 1500 = 0.13%. PPPoE imposes an additional 6 bytes, giving a total of 8 bytes as per my original response. I understand why there is so much conjecture; it is a very complex topic with many layers and protocols all interacting in various ways with each other.
Here is an excellent paper that I think does a superb job of explaining this whole topic:
That discussion thread is full of inaccuracies and incorrect terminology and comparing apples with oranges (the opening sentence of the original post "we know that PPPoE overhead..." is completely wrong at multiple levels).
For example, the overhead of AAL5 (which sits on top of ATM), varies between 8 bytes and 55 bytes per IP packet transmitted, and the size of an IP packet is variable, so right there you see a slight problem with working out a percentage overhead, when both the overhead and the payload sizes vary as the payload has to be fit into 48-byte ATM frames.
Strictly speaking, the discussion was about the overheads of PPP, and PPPoA imposes a 2 byte overhead per IP packet, that is a fact. 2 / 1500 = 0.13%. PPPoE imposes an additional 6 bytes, giving a total of 8 bytes as per my original response. I understand why there is so much conjecture; it is a very complex topic with many layers and protocols all interacting in various ways with each other.
Here is an excellent paper that I think does a superb job of explaining this whole topic:
Yes, Thanks for providing an additional link with more information. It's a very interesting topic that we are all learning from and I also appreciate the information that is coming out of this discussion. These kinds of router and connection issues are the kind that many people will have to deal with from ISPs especially as internet bandwidth increases and usage caps are implemented. I wasn't scientifically verifying the actual percentage loss of using PPPoE with DSL that was discussed in those posts. The reason I provided that link is that it represents the kind of posts and threads I have been reading for more than a decade in the forums that discuss this same topic about losses in throughput.
This has never been a hidden secret. The ISPs have known about it and the techs talk about it. I don't claim to be an expert on PPPoE nor do I want to imply that I know much more about it than someone else in this forum or any other forum who seems to have more knowledge about it. However, it is something I have asked former DSL ISP sales reps and techs on the phone about in the past in regards to their internet service and they have admitted that losses exist. The precise reasons for the losses, lost packets, when using PPPoE can be analyzed in detail but there definitely are significant verifiable losses that most users experience on DSL lines using it. You said you measured 300Mbps down from your service so it could be that your fiber ISP is compensating for losses with additional bandwidth or maybe there is another factor with how your provider is implementing it so you don't notice the impact of using PPPoE with their equipment.
Last edited by SoCalReviews; 01-28-2013 at 04:40 PM.
Would the new EdgeRouter Lite from Ubiquiti be an option for you here? It appears that it does support PPPoE, though it doesn't appear to have the most user-friendly configuration... But if you're comfortable with Mikrotik's RouterOS, I'd imagine you could plow through their implementation of Vyatta with minimal hassle. Plus it's only $99 USD.
Personally, I would pitch in for an atom barebone from newegg, throw in pfsense for firewall/routing (or build Openbsd based unit for fun), use the N66U strictly as a switch & AP behind it. You get enterprise class throughput, shaping and routing with pf stack.
I believe that the RT-N56U's hardware NAT processor can offload PPPoE. Also, I am curious if the modem is properly bridged, and the router is handling the packet properly?
It seems as if there is fragmentation or a path MTU leading to fragmentatio, and this is causing the punitive slow down. I could very well be wrong, but I would do just a little more investigating before the final decision.
Would the new EdgeRouter Lite from Ubiquiti be an option for you here?
It's possible, but they are very new and only just starting to trickle into the marketplace, so will have to wait and see what the reports are. And from experience, marketing and CPU speed don't correlate to PPPoE performance. Also, even though they are "just $99", here in the UK they are going for around £115 (almost $180).
I believe that the RT-N56U's hardware NAT processor can offload PPPoE. Also, I am curious if the modem is properly bridged, and the router is handling the packet properly?
It seems as if there is fragmentation or a path MTU leading to fragmentatio, and this is causing the punitive slow down. I could very well be wrong, but I would do just a little more investigating before the final decision.
Unfortunately, I have been unsuccessful in unearthing any evidence to support your belief.
Unfortunately, I have been unsuccessful in unearthing any evidence to support your belief.
With no real technical information of your network, I can say that it would be the case for us too. But to note, that most issues with networking are simple 101 problems. Never really going beyond that much. Personally, from my DSL PPPoE days, I never experienced low throughput and bandwidth issues; and this was with less capable networking hardware of the day.
So, the curious question is, is your modem bridged, and only bridged? Therefore passing on the packets to the network (router). Is the N66u handling the PPPoE, and only it? I do recall a website that when testing the N66U the PPPoE bandwidth was higher than 500Mbp/s. I cannot find it at the moment, but that router should be more than capable of handling that. Not all the packet processing is handled by the processor with this router; it does have some offloading abilities too. Not to the extent of the N56U.
Linear Mode
