I had a briefing call with one of the mobile video optimisation companies yesterday, discussing the role played by network boxes that can compress, rate-limit and otherwise change a downloaded video stream.
It's an area I've been critical of for a while - especially where the network 'transparently' alters video content, without the explicit consent of either video publisher or end-user. I recognise that's a useful short-term fix for congested networks as it can shave 20-30% off of data throughput, but I also think that it's not a viable model in the long term.
Basically, publishers don't like the idea that some device in the data path "messes with our content", and in many cases users and regulators won't like it either. Some of the vendors suggest that reducing video "stalling" by optimisation actually improves the quality of user experience (QoE), but there are also other approaches to that which are more "consensual".
It was interesting to read the other day that Verizon's Video arm (formerly V-Cast) is working with its network teams to pre-define properly "optimal" formats for video depending on network connection, device type (ie screen & processor) and even user dataplan. In my mind, that's totally acceptable, as the content publisher/aggregator is working hand-in-hand with the network team to create a balance between QoE, network integrity and "artistic integrity". Collectively, they've thought about the trade-offs for both network and user in terms of quality, cost and performance. Presumably they also looked at the role of CDNs, adaptive bitrate streaming, on-device caching and assorted other clever mobile-video tech. According to a related article, they can't use WiFi offload because their content rights agreement is for 3G/4G only - another good illustration of the complexities here. All in all, it sounds like a great example of "holistic" traffic management, working with the various constraints imposed. I assume they have a roadmap for further evolution as it all matures.
But there's a million miles between that philosophy, and the type of non-consensual optimisation that's increasingly common for normal Internet video traffic destined for mobile devices, where the network unilaterally decides to alter the content. We're seeing solutions getting a bit smarter here - for example, only acting if the cell is actually busy, but there are various other use cases where it's more directly about the operator's costs.
The vendor I spoke to yesterday mentioned scenarios like shared networks (where each operator pays a share of costs based on traffic volumes), or where the backhaul is obtained from a third-party fixed operator at variable-cost. Another scenario I've heard (usually more for caching / CDNs) is around international transit in parts of the world with expensive connectivity. But another use cases was where the network is uncongested but high-quality video was being downloaded at high speed. In that instance "we can take 20% of the data stream off the top, and they won't notice".
Here is where we differ. In my view, if I (as a customer) pay for Internet Access, then I expect that the bits and bytes that come out of the server are the same bits and bytes that arrive on my device. The server owner thinks the same. Yes, in certain circumstances I'll accept a trade-off if it improves my QoE, as long as I am told what's happening and opt in: imagine an icon appearing, indicating "optimiser on", or a switchable user-controlled optimiser like Onavo's for reducing roaming traffic. But I'm not prepared to accept a lower quality just because the operator has been stupid enough to agree contracts for a shared network, with terms that don't take account of the types of retail broadband service its selling to its customers.
"Oh, I'm sorry Mr Bubley, we know you paid £1000 for the flight, but we've downgraded you to economy class because it's a codeshare flight, and we have to pay our partner airline £500 extra per-seat for business class passengers". I'm sorry, but that's your problem and not mine.
And as for the notion that the Internet connection can arbitrarily decide that "I won't notice" if it chops up & reformats my data? Hello? How do you know that? Maybe I've encoded something into it steganographically in the background? Maybe I've PAID for content of a specific resolution, and I have a contract with the publisher? Maybe I'm willing to take the trade-off of throttled buffered' throughput, because it's going to a background app for viewing later? If I've paid for my 2GB per month, I'll use it the way I like, thank you very much. If the network's busy, fine - tell me (and/or the publisher) & I'll use WiFi or download it later or *I* will make the decision to drop the quality if I want instant gratification.
I've got a lot of sympathy for operators that have genuine peaks and troughs of demand, or that are constrained in terms of supply by spectrum shortages or difficulties obtaining cell sites. But I've got less sympathy for companies that sell a product ("2GB of mobile Internet access at up to 10Mbit/s") and then realise they can't deliver it, because it doesn't match up to their network's capabilities or cost structure.
But it struck me that the reason that telecom operators (especially mobile) think that this is OK, is that degrading quality is at the core of their business proposition. For the last 100 years or so, we've had to deal with the fact that our natural, wideband human speech has been squished into 3KHz pipes for transport across the network. It's "optimised" (ie downgraded) at the microphone on your handset. Of course, now we have service providers trying to monetise HD voice - and expecting you to pay a premium just for transporting what your vocal cords have been putting out all along.
I'd love to know if there's internal transfer-pricing going on at Verizon for the video service. Is VZW actually "monetising" its optimisation capabilities and charging the Video department for the privilege? Or is it being done for mutual benefit, without money changing hands? I suspect the latter, which is the model I see most (but maybe not all) consensual video optimisation working out.
Oh, and if I ever go for a drink with this particular optimisation company exec, I'm going to have a word with the barman. I'll tell him to pour 80% of his pint of beer as normal, but top up the other 20% with coloured water. I'm sure he won't notice.
It's an area I've been critical of for a while - especially where the network 'transparently' alters video content, without the explicit consent of either video publisher or end-user. I recognise that's a useful short-term fix for congested networks as it can shave 20-30% off of data throughput, but I also think that it's not a viable model in the long term.
Basically, publishers don't like the idea that some device in the data path "messes with our content", and in many cases users and regulators won't like it either. Some of the vendors suggest that reducing video "stalling" by optimisation actually improves the quality of user experience (QoE), but there are also other approaches to that which are more "consensual".
It was interesting to read the other day that Verizon's Video arm (formerly V-Cast) is working with its network teams to pre-define properly "optimal" formats for video depending on network connection, device type (ie screen & processor) and even user dataplan. In my mind, that's totally acceptable, as the content publisher/aggregator is working hand-in-hand with the network team to create a balance between QoE, network integrity and "artistic integrity". Collectively, they've thought about the trade-offs for both network and user in terms of quality, cost and performance. Presumably they also looked at the role of CDNs, adaptive bitrate streaming, on-device caching and assorted other clever mobile-video tech. According to a related article, they can't use WiFi offload because their content rights agreement is for 3G/4G only - another good illustration of the complexities here. All in all, it sounds like a great example of "holistic" traffic management, working with the various constraints imposed. I assume they have a roadmap for further evolution as it all matures.
But there's a million miles between that philosophy, and the type of non-consensual optimisation that's increasingly common for normal Internet video traffic destined for mobile devices, where the network unilaterally decides to alter the content. We're seeing solutions getting a bit smarter here - for example, only acting if the cell is actually busy, but there are various other use cases where it's more directly about the operator's costs.
The vendor I spoke to yesterday mentioned scenarios like shared networks (where each operator pays a share of costs based on traffic volumes), or where the backhaul is obtained from a third-party fixed operator at variable-cost. Another scenario I've heard (usually more for caching / CDNs) is around international transit in parts of the world with expensive connectivity. But another use cases was where the network is uncongested but high-quality video was being downloaded at high speed. In that instance "we can take 20% of the data stream off the top, and they won't notice".
Here is where we differ. In my view, if I (as a customer) pay for Internet Access, then I expect that the bits and bytes that come out of the server are the same bits and bytes that arrive on my device. The server owner thinks the same. Yes, in certain circumstances I'll accept a trade-off if it improves my QoE, as long as I am told what's happening and opt in: imagine an icon appearing, indicating "optimiser on", or a switchable user-controlled optimiser like Onavo's for reducing roaming traffic. But I'm not prepared to accept a lower quality just because the operator has been stupid enough to agree contracts for a shared network, with terms that don't take account of the types of retail broadband service its selling to its customers.
"Oh, I'm sorry Mr Bubley, we know you paid £1000 for the flight, but we've downgraded you to economy class because it's a codeshare flight, and we have to pay our partner airline £500 extra per-seat for business class passengers". I'm sorry, but that's your problem and not mine.
And as for the notion that the Internet connection can arbitrarily decide that "I won't notice" if it chops up & reformats my data? Hello? How do you know that? Maybe I've encoded something into it steganographically in the background? Maybe I've PAID for content of a specific resolution, and I have a contract with the publisher? Maybe I'm willing to take the trade-off of throttled buffered' throughput, because it's going to a background app for viewing later? If I've paid for my 2GB per month, I'll use it the way I like, thank you very much. If the network's busy, fine - tell me (and/or the publisher) & I'll use WiFi or download it later or *I* will make the decision to drop the quality if I want instant gratification.
I've got a lot of sympathy for operators that have genuine peaks and troughs of demand, or that are constrained in terms of supply by spectrum shortages or difficulties obtaining cell sites. But I've got less sympathy for companies that sell a product ("2GB of mobile Internet access at up to 10Mbit/s") and then realise they can't deliver it, because it doesn't match up to their network's capabilities or cost structure.
But it struck me that the reason that telecom operators (especially mobile) think that this is OK, is that degrading quality is at the core of their business proposition. For the last 100 years or so, we've had to deal with the fact that our natural, wideband human speech has been squished into 3KHz pipes for transport across the network. It's "optimised" (ie downgraded) at the microphone on your handset. Of course, now we have service providers trying to monetise HD voice - and expecting you to pay a premium just for transporting what your vocal cords have been putting out all along.
I'd love to know if there's internal transfer-pricing going on at Verizon for the video service. Is VZW actually "monetising" its optimisation capabilities and charging the Video department for the privilege? Or is it being done for mutual benefit, without money changing hands? I suspect the latter, which is the model I see most (but maybe not all) consensual video optimisation working out.
Oh, and if I ever go for a drink with this particular optimisation company exec, I'm going to have a word with the barman. I'll tell him to pour 80% of his pint of beer as normal, but top up the other 20% with coloured water. I'm sure he won't notice.
Hi Dean,
ReplyDeleteA very interesting view, however there is no black & white here.
As you mentioned video optimization has two benefits:
1. It reduces the load on operators networks
2. It improves the subscriber QoE both temporal quality (reducing the amount of stalls and speeding video start time) and Visual quality.
I totally agree with you that subscribers should be aware of what they signed for and the example you gave from VzW and what we heard from other operators we know show that this is exactly what they have in mind.
If you pay more you get more (including enhanced QoE through optimization), If you pay less you get less.
Using your analogy to airlines:
Today operators are selling all tickets at the same price. If you got on the plane first you get to seat in first class, if you got on last you get the middle seat in front of the toilets.
When operators will tier their offering they will sell First class tickets, Business class tickets, Tourist class, isle seat and exit seat tickets.
You will then get exactly what you paid for
Hi Tal
ReplyDeleteI disagree that network optimisation necessarily improves QoE - it can affect a slightly wider slice of QoS, but true end-to-end QoE cannot be measured or "optimised" purely in the network.
To do that, you need a software client either directly on the device, or better integration with the origination source. Otherwise, at best, you are second-guessing the user's actual context. For example, stalling is irrelevant for a video downloaded to a background window for viewing later, and lossy compression makes the ultimate user experience worse when it is actually watched.
Airline analogies are dangerous, given the number that have gone bust in recent years.
Many operators are already offering tiers (eg by speed). Tiering by quality / VIP status is feasible but needs near-perfect coverage and better capacity in order to be meaningful.
Also, look at the Ofcom announcement today, which points the right way. Normal "Internet Access" will be neutral and subject to the operator stipulating average speeds, but they can also offer "Managed Services" as well.
Dean
I so hated it when I was downloading for professional work use images from stock photo sites that I had paid for, and T-Mobile's mobile broadband would decide to make them look awful.
ReplyDelete"In my mind, that's totally acceptable, as the content publisher/aggregator is working hand-in-hand with the network team to create a balance between QoE, ..."
ReplyDeleteI did not read in the news that Verizon is working hand-in-hand with content publishers. It seems they are taking the decision of downgrading video quality without involving the content provider.
Where did you read it?
Davide
ReplyDeleteI'm referring to the specific example of Verizon's *own* video streaming service called Verizon Video (formerly Vcast).
Based on the Fierce Wireless article, it looks like the content division of VZ is working with the network division of VZW.
However, I also understand that VZW uses "non-consensual" optimisation for content from other sources. But at least it makes that fact clear to customers. I'm not certain if it does it across-the-board or just when it's actually experiencing congestion.
That approach is effective as a common short-term sticking plaster, but is likely to get pushback over time, either by publishers encrypting their content, or by using a quality indicator to alert users when the carrier is "messing about with their data".
Dean you write:
ReplyDelete"I disagree that network optimisation necessarily improves QoE - it can affect a slightly wider slice of QoS, but true end-to-end QoE cannot be measured or "optimised" purely in the network."
I concur when a specific optimisation is used. For example, if you throttle back P2P then streaming video may take more bandwidth resulting in an equivalent congestion problem. The other way around is also true, optimise video the P2P will eat what is left. This is because streaming video and P2P are greedy protocols.
Regardless of what networks do now in early days (yes I still think we are in the early days), the eventual solution needs to be a comprehensive approach to congestion management.
"Optimisation" should be stricken from the vocabulary for fast HSPA and 4G, since it really does not benefit them much. Speeds are fine when congestion is not occurring.
Reducing the bandwidth for a video just like throttling P2P is only necessary when congestion is occurring and only makes sense if you can control all the data going to the congestion point.
At the end of the day when congestion occurs something happens. The question to the carrier is do they want to control that (e.g. give the gold user a better experience because they pay more) or just let the Internet protocols figure it out. Most carriers will want the former and not the later and I agree with that opinion.
Taking this to the next level, when congestion is not occurring carriers should try to get applications to use more bandwidth. The VZW announcement is a step in the right direction. I'm sure there are others as well that will be announced.
Carriers have to provide the best QoE with the network they have. How they choose the "best QoE" will undoubtedly differ, but they again I think it should. If a carrier makes a mistake (e.g. block Skype) the market will reach and correct them. If they do something smart, sensible, and transparent (e.g. you pay for first class on a plane you will be treated differently) then they will be rewarded.
Anonymous
ReplyDelete"Carriers have to provide the best QoE with the network they have"
I'd modify that statement to:
"Carriers have to provide the best QoE with the network, software, influence, content/apps, partners & devices they have"
The best way to avoid being "just a pipe" is to make sure you aren't just in the business of running pipes (ie networks).
The "best QoE" will come from companies that match the best pipes to the best taps/valves (eg phones or on-device software), the best water (content & apps), the best partnerships (waste management etc) and so on. Maybe even the best "water experience" by sponsoring the London 2012 Olympic Swimming.
I also think the "first class" plane analogy is deeply flawed, especially on mobile networks rather than fixed. In mobile, first class is ALREADY awarded to people with the best radio conditions: that's how HSPA gets to be 28.8Mbit/s, or LTE to 100Mbit/s.
The moment you start fiddling about with the packet prioritisation at the base station is the moment you get into hot water (pun intended, sorry) with the consumer-protection lawyers.