I thought that SMPTE 2022-6 was a joke, but then came SMPTE 2110. For anyone wanting to do virtualization, cloud or standard IT-based implementations this is a slap in the face for any software engineer. Old school hardware engineering combined with design by committee at its best. This can't be salvaged.
BROADCAST VIDEO OVER IP "STANDARDS"
Let's add some perspective and have a look at the state of the industry right now and try to make sense of the many acronyms thrown around. There are a number of seemingly competing standards. Let's look at the major contenders.
● ‘Simple DVB‘ / ‘MPEG2 TS’
● Understood by many, supported today
● Low rate
● Complex to manage data events
● Codec-dependent Latency
● Many software implementations
● Does work in the cloud just fine
● ‘SDI over IP’
● All blanking ‘kludges’ carry over
● Requires high perf NICs and switches
● Strict jitter / latency expectations
● Supported and working now
● No software implementation
● Does not work in the cloud
SMPTE 2110 / AES-67
● ‘Uncompressed elementary streams’
● Large data rate (>1Gbps for HD)
● ANC data transcription (CC, AFD)
● Requires high-perf NICs and switches
● Not self-described or discoverable
● Poor scaling to UHD ( >8Gbps !)
● Needs NMOS to do anything useful
● No software implementation
● Does not work in the cloud
● Metadata layering for 2110 / AES 67
● Discovery for 2022-6 / 2110 / AES 67
● RESTful API for queries and registration
● Requires mDNS / Bonjour
● No security (at all, by design)
● Metadata layer standard is controversial
● ALLIANCE FOR IP MEDIA SOLUTIONS
● No technical standard or work
● Loads of stickers and lapel pins
● Propaganda trade org to sell 2110
● Seemingly dead
● Like 2110 / AES67, but uses MPEG2 TS
● Self-describing, elementary
● Gains benefits of MPEG2 TS experience
● Large data rate (1Gbps for HD)
● Requires high perf NICs and switches
● All ANC signals transcribed (CC, AFD, etc.)
● Low-latency, compressed streams
● Self-describing, simple discovery
● Modest data rate (100Mbps for HD)
● Designed for software use
● Ready to go, free SDK
● Proprietary, black box
● No low-level API or QoS control
● Enterprise scalability unproven
TICO OVER IP
● TICO now also with misleading "IP" label
● SDI in-line 4:1 compression or
● 2022-6 / 2110 4:1 in-line compression
● No TICO IP stream exists as such
● Adds "compression" to SDI/2202-6/2110
● All other problems remain the same
SONY IP LIVE
● Seemingly dead
● Sony seems to be AIMS / 2110 now
● Hardware only
● Compressed - multiple UHD in 10GbE
● Proprietary, black box
Studying the tables above, it becomes clear that not one single Video over IP standard does it all. But some are worse than others, especially when your vision is to have a legacy free software-only based environment, where the limit of what you can do with video, is only defined by the available computing power, as well as networking and storage bandwidth and resources - either local, virtualized or in the cloud.
THE SMPTE 2110 PROBLEM
SMPTE 2110 with a limited usage scenario e.g. inside a studio would be fine, although where the benefits over SDI would be is hard to see. This is a general problem for most broadcast video over IP solutions - as there are virtually no cameras that support the respective IP standards, SDI camera outputs still needs to be converted to IP and back to SDI again e.g. to connect a monitor, vision mixer or anything else - and that all over the place. The vendors selling converters or "gateways" stand to make the most money.
All this could be chalked of as birthing pains, but now a large part if the industry is telling broadcasters that SMPTE 2110 is the one solution that can do it all and that you better join the bandwagon if you do not want to be left behind.
There are many good reasons why everyone should sit this one out, even one has money to burn and feels the urge with UHD and 8K one needs to do infrastructure investments now.
Don't - at least not on SMPTE 2110.
WHY SMPTE 2110 IS AN UNFIXABLE TRAIN WRECK
Where to start? Maybe by saying it is half-baked, unfinished, was pushed out of the door before all problems were addressed that now need to be fixed by NMOS side car projects? No, that would assume it could be fixed eventually.
The whole approach to "solving" the problem of video over IP shows where the actual problem lies - the type of people that were involved solving it.
If you ask broadcast hardware engineers to solve video over IP then you get SMPTE 2022-6, which is SDI in its entirety wrapped into an IP stream. To create this in software you need to emulate a complete SDI video card and all the legacy "features" of the SDI standard as well. Of course this is uncompressed and at least 1.5Gbps for a simple 1080i video stream.
After the backlash received for creating SMPTE 2022-6, which was essentially still SDI, just swapping the transport medium - continuous serial stream to packetized Ethernet, SMPTE 2110 now is almost the other extreme. Now everything is completely unpacked and stuffed into separate, undescribed elementary RTP streams - creating a whole range of new problems - and of course it is also uncompressed.
Why uncompressed? There theoretically may be a case for uncompressed. Theoretically. This stupid discussion resurfaces with each and every technology change. Bring back LaserDisc! No (affordable) tape format since analog Betacam was uncompressed. Yes, D1 and D6 were, but apart from some broadcast labs no one could afford them and no one used them in production. DigiBeta, HDCAM / HDCAM SR etc. are all using compression and still no one ever complained or said the quality was not good enough for high-end post. In cinema production most cinematic "RAW" formats use compression and many prefer shooting in high-quality ProRes or DNxHR directly to save space and production time. Many statements about uncompressed are myths kept alive by propagating lies, like saying that uncompressed has lower latency or is needed to be able to do a proper chroma key. But even the question of uncompressed or not, is not the real problem, it is just creating unnecessarily large data, which severely limits the usage scenarios where SMPTE 2110 can be used.
THE REAL KICKER
The biggest kicker about SMPTE 2110 is the fact, that with its latency and jitter requirements, it is almost impossible to implement in a pure software stack. One that could run for instance in a popular virtualization environment like VMWARE or Hyper-V. For now the only way to get it to work on a standard IT server requires a physical card from vendors such as AJA or Deltacast, if these were already released yet, and working. Neither of these cards do not support UHD, so if UHD is what you want, then search on. As an alternative one could use a high-end 25G Ethernet adapter with complete hardware offloading and dedicated drivers, which is in the end not much different from having dedicated video I/O card.
In a VM environment any of these cards would have to be either passed-thru directly - one card per VM or special, VM-version specific drivers would have to exist. For the Mellanox 25G/50G/100G ConnecttX-5 card some of these condition are met already e.g. for VMware.
But in the end all of this currently offers little benefit over SDI, which for now remains considerably cheaper. Making a pure software based SMPTE 2110 stack will be possible eventually, but only using very specific high-end enterprise 10/25/50/100G NICs and some well written drivers. This will not work what is generally found on a PC or server motherboard already.
A real-time OS would help to achieve this with simple NICs - but this would then exclude using Windows, MacOS, standard Linux, and virtualization/cloud.
THE REAL KICKER - PART2
SMPTE 2110 on its own does not really work. Its wonderful elementary nature splits it into video (2110-20), audio (2110-30) and ANC data (2110-40) and a separate time "stream" (2110-10 / SMPTE 2059) basically using PTP driven network timing. None of this tells you what you are dealing with. Without the Session Description Protocol providing a "text file" with the definition of what is really in those streams, they are meaningless raw data. The SDP and how this is controlled is not a part of the scope of SMPTE 2110. This is where 2110 needs to enlist the help of AMWA / BBC R&D's NMOS which is the magic bullet, which will fix this all. Or not. As matter of fact getting different vendors devices to talk to each other is a bit of an adventure in real life. Different vendors have extended the protocol to their liking or interpret SDP differently. Then there is still the issue of security and large scale discovery. Or the question of who calls the shots should you ever entertain the crazy idea to connect two SMPTE 2110 equipped OB trucks to each other. The only solution? Go SDI or 2022-6!
SMPTE 2110 Uncompressed Bandwidth – the Madness in Numbers
Let’s do a sample calculation.
Setting a simple goal – moving a small TV station from HD to UHD - 1.5G/3G SDI 1080 to UHD @ 59.94 fps. In the process an existing 3G SDI 64x64 router is going to be replaced by Ethernet switching hardware.
Using SMPTE 2110 uncompressed this means the UHD data does not fit inside a single 10Gbit Ethernet link, further assuming that the UHD is one video stream that is not split across four HD streams, then trunking two 10GbE connections is not going to work either. This means using at least 25GbE per UHD stream. For SMPTE 2110 and UHD this is going to be the norm.
Theoretically you could squeeze two UHD@59.94 streams into a single 25GbE connection, but if the source is a physical device such as a camera, vision mixer, etc., then aggregating two streams into onw uplink will only be possible by using yet another local Ethernet switch.
Multi-path resilience is a major selling point going from SDI to IP and anyone architecting a SDI to IP replacement solution will factor this in. This translates into two 25 GbE connections per UHD signal and two physically separate Ethernet switches either link goes to. The goal being to replace a 3G SDI 64x64 router we now have the following numbers: two times 64x 25 GbE ports – split over at least two physical Ethernet switches.
There are no standalone 64x port 25GbE Ethernet switches on the market currently. There are 32x port and 48x port 25GbE switches (with multiple 100 GbE uplinks), but the only way to get 64x 25GbE ports in one box is to get e.g. a Cisco Nexus 9500-series modular switch platform which can be configured with various modules. The smallest Nexus 9500 box with four empty slots is the Nexus 9504 which needs either two or four 48x port 1/10/25GbE line cards (9K-X97160YC-EX) added. Two cards makes 2x 48 = 96 25GbE ports – so this would be one redundant Ethernet leg, but for full redundancy a second Nexus 9504 unit with also two 48x port line cards is needed.
If the opinion is that redundancy is for sissies one Nexus 9504 with four cards will do, of course. But then one can also do away with the whole multipath redundancy and cut the number of ports required by half.
Knowing broadcasters, corners will be cut, especially when they see how much top notch network equipment costs. “We did not have redundant SDI routers either and never had a problem …”, will be the often-heard statement. “Lucky you!”, is the right answer, followed by “of course you are right. Let’s migrate to a new, barely working IP video standard that we have no experience with, and do so without having another leg to stand on. Redundancy is for losers.” That will save the organization a lot of money to start with. This should go hand-in-hand with updating one’s CV, Monster profile and LinkedIn contact details.
Maybe sticking with a redundant design is the better alternative especially when spending other people’s money. A single Cisco Nexus 9504 with controller, four fabric modules, and two 48x port line card is going to set you or your employer back around $80k – times two for redundancy – equals around $160k.
Not everyone will need/want to pay for Cisco, so doing this with two Mellanox 48x 25GbE / 8x 100 GbE Mellanox MSN2410 switches per redundant leg would be a little over half the price of the Cisco solution. This is still around $100k to replace a 3G SDI router that depending on the vendor and scalability options starting from less than $10k (e.g. a AJA Kumo 6464).
The power consumption of each Cisco 48x 25GbE line card is 12x as high as that of a complete AJA Kumo 6464 3G router and at least two line cards are needed, plus controller, fabric modules, chassis etc. Around 1kW is the expected power consumption for each Nexus 9504 unit (with just two line cards). That is more than 10W per 25GbE port. The Mellanox 25/100G switches are less power hungry consuming only 5x as much power as the AJA Kumo 6464 – each, or less than 3W per 25 GbE port.
Yes, this is petty and it is comparing apples with pears, but this stands to show that just by saying “it is IP” makes it automatically good or clever or efficient. This is not to say IP is bad, but using compression – be it lossless, visually lossless, or slightly lossy, then the whole IP UHD workflow could be handled using 10 GbE connections or even ordinary Gigabit Ethernet. Depending on the compression level chosen even multiple UHD streams could fit into a single 10 GbE link.
The resulting cost and power savings are dramatic. 10G Ethernet switches are absolute commodity and priced accordingly. Gigabit switches with 10G uplinks are laughably cheap making the time spent on comparing them quickly more expensive than buying them.
THE WINNER IS ... NDI?
It comes at little surprise that even large broadcasters now start dabbling with Newtek's NDI, which has its own set of problems and limitations, but also a fair amount of advantages. It has a lot of plus points; it is an easily deployable software-only stack that runs on older hardware with non-enterprise NICs just fine, it unashamedly uses compression and therefore can do multiple HD streams over simple GbE, it works in VMs and has a simple, free SDK and sample apps to go with it - all this makes it gain more and more followers every day. NDI is a proprietary blackbox that is very simply configured because there is little or nothing to configure. For broadcast engineers the lack of low-level control and QoS monitoring is a nightmare as is the completely opaque nature of the codec used inside. Enterprise wide discovery and scalability are still work in progress, although it is progressing. The advantage of NDI's proprietary nature is that Newtek can drive its development at much faster pace than the "competition" stuck in SMPTE or VSF committees having to agree on the lowest common denominator.
THE WINNER IS ... REALLY?
No one for now. Either Newtek opens up NDI for real or a new 3rd option comes along. The SMPTE 2110 folks talk about adding compression to the standard, but that is likely to be just something like TICO, which really does not help anyone.
THE REAL WINNER COULD BE SMPTE 2022-2
Maybe the answer already exists - why not just use SMPTE 2022-2 but with much higher bit rates or other higher quality codecs. Creating a Transport Stream with AVC-Intra 100 or AVC-Ultra inside. This would be self describing, low latency, can handle all the ANC data and would work with most existing tools out there already. VLC can even already play these streams out of the box.
TIME WILL TELL
Chances are that SMPTE 2110 could still win as the forces of marketing and fake news are unstoppable. Also, the fact that technical savvy broadcast journalists are an extinct species having been replaced largely by copy & paste bots, makes this a likely scenario.
The only blessing in disguise is, that the broadcast industry as a whole is becoming more and more a small sideshow to the Internet / Telco industry that has assumed ownership of what we may still call television. This might bring on massive changes - if the lessons learned in these industries are being applied - or not, if the new owners consider the margins high enough for now and the rest will get addressed by attrition.