LPT Inflation vs. ETH Rewards from transcoding

As an orchestrator/transcoder that is currently only doing CPU transcoding, I wanted to share my thoughts regarding yesterdays discussion in the community call.

Maybe a short recap: The LPT inflation is currently decreasing at a rate of 0.03 percentage points per round and will reach 0 in less than 7 months if nothing is changed. By then, an orchestrator’s revenue is solely dependent on the ETH rewards from its transcoding jobs.

One opinion is that this will be good for the network since it weeds out the orchestrators that don’t provide proper transcoding (which is what livepeer is all about in the end). One of those transcoders would be me at the moment :slight_smile:

I would love to help the network and provide some GPU transcoding, but as someone that doesn’t already have a mining rig, it just makes zero sense economically. I’m living in a flat, so setting up a few GPUs is currently not an option due to heat/noise. Colocation would be an option but for that to pay itself, the ETH rewards would need to be quite substantial first.

IMO, livepeer needs more supported devices, e.g. Intel QuickSync or AMDs Ryzen with integrated Vega GPU - that would at least give people like me the option to set up a small mini PC for some light transcoding while the ETH fees grow and a bigger investment becomes reasonable.
But what I would really love to see is support for devices like the nvidia Jetson Nano - a low cost, low power, passive cooled device that is capable to do 4k transcoding! This is something that almost everybody can afford and run and would allow Livepeer to grow a truly decentralized transcoding network all around the world. I for one would be happy to run a few of those (I do have a reliable 1 Gbit/sec up- & download connection).

Now let’s assume we have 0 inflation in 7 months but only a little bit of fees are generated. For many orchestrators it would not be worth it to invest in a transcoding setup. So most of the bonded LPT (also I would expect a large sell-off…) would end up with a few orchestrators that already have the setup (maybe because they’re also mining).
And once a good amount of ETH fees are generated, I see a chicken/egg problem: A new transcoder (that joins the network in maybe one or two years) does not get enough work because he does not have enough stake (because nobody will bond to him because he does not generate enough ETH fees…). So essentially he’s forced to “work” with an orchestrator that already receives transcoding jobs (and that can dictate the terms)?

Only a low number of supported devices, 0 LPT inflation and low LPT liquidity would increase this “centralization” of orchestrators.

So my opinion is to keep a healthy amount of LPT inflation until enough ETH fees are generated. In the short term, this might keep some “lazy” orchestrators running that don’t care about transcoding at all, but in the long term Livepeer would end up with a more decentralized set of orchestrators.

2 Likes

support for devices like the [nvidia Jetson Nano---- a low cost, low power, passive cooled device that is capable to do 4k transcoding!

I don’t think you need to acquire a full rig.

I acquired a 2 GPU rig using 2nd hand GPUs (from a gaming team so no mining or whatever done on them) for less than 1000$. Some googling allowed me to uncap the session limit on consumer grade nvidia GPUs for encoding video.

The amount of concurrent streams I can transcode should be sufficient for the first couple of months of real demand into the network if other operators do the same thing.

Nontheless I’m in the same boat as you. I live in an appartment, power is expensive here and bandwith and fixed IPs isn’t the cheapest either (thank god I can run my remote transcoder on a dynamic IP and run my O from a datacenter). Therefore I searched for other ways to gain a competitive advantage over operators that do have the necessary resources to fund large operations and I came up with Livepool.

Creating a public transcoding pool is challenging , even when one has good knowledge of the livepeer codebase but perhaps there’s other things you can do such as work with mining rig operators on a contract basis and they have to meet certain SLA Requirements. That’s much easier than trying to run a public pool without SLA guarantees.


On-topic:

I did like @Gavin 's proposal during the call. An LPT inflation curve that is based on the amount of fees flowing through the network. I’d have to look at the protocol but this maybe even be an implementable option (or a combination of participation and then providing a base inflation rate if there aren’t enough fees on the network).

@vires-in-numeris

But what I would really love to see is support for devices like the nvidia Jetson Nano - a low cost, low power, passive cooled device that is capable to do 4k transcoding!

This is the finest idea I have heard in Livepeer Community for a looong time! The ability for each of us to add our own physical hardware to transcode video feels like an ideal way to build the network in a decentralised permissionless way, whilst dogfooding things for ourselves.

This is especially in a context where DAppNode, Ethereum, and other projects are looking to arm as an alternative to amd64, as a way to empower self-hosting instead of “just run it on Amazon”.

But in some ways, there is a question of prioritisation between “building for big mining farms to get volume” vs. “building for small independent startup operations for get decentralisation”. I expect that Inc. might tend towards the former given their current plans for bringing on scale.

Perhaps then this is something for the community to pick up, perhaps with backing from some community grants? After all, the community is the pure decentralised / permissionless part of the project… perhaps building on the work being done by videoDAC with Livepeer / ARM

@NicoV

I acquired a 2 GPU rig using 2nd hand GPUs (from a gaming team so no mining or whatever done on them) for less than 1000$.

At that price point, I would need to do a bit more research to feel confident I would know what I’m doing. Plus, the awareness of whether the exact custom hardware I am buying is compatible with this nascent tech. Also, a rig sounds unpleasant. Better to get a nice case and it can sit under the TV (or better, inside the TV!).

If these things could come with $100 price tag, and be “certified” to work (whatever that means), perhaps even with custom OS image build - so you can just write the image to the SD card, plug it in, turn it on, create your keys, then Invoke multi-step "become an orchestrator". I’d get one for my Dad.

They could even run in Transcoder-only mode, just select which Orchestrator you want to contribute your GPU cycles to (for love, not money!), then put in an -orchSecret.

There was some good discussion on Discord recently about the pros/cons of adjusting the LPT inflation schedule. I’ve included the chat logs below for reference.

Discord Chat Logs

[15-May-20 07:26 PM] vires-in-numeris#5324
I actually love the idea of tying the LPT inflation to the amount of fees generated!
It allows the orchestrators/transcoders to keep operating long enough until an investment in proper transcoding gear is actually worth it. Whereas those orchestrators who already have a setup, would profit from the additional ETH fees as well as probably more LPT rewards (since more people will bond to them due to the fees generated).
I also shared my thoughs regarding this topic (before I read about this idea…) here: LPT Inflation vs. ETH Rewards from transcoding

[16-May-20 04:41 AM] f1l1b0x#6391
the idea that someone buys dedicated hardware to participate as a new O is fundamentally flawed. The expectation that a protocol should pay you for your investment up front is flawed as well. I am more and more against a extension of the inflation as it will just pay those who already got tons of token for not transcoding but having a rpi staking.

[16-May-20 06:14 AM] vires-in-numeris#5324

the idea that someone buys dedicated hardware to participate as a new O is fundamentally flawed. The expectation that a protocol should pay you for your investment up front is flawed as well. I am more and more against a extension of the inflation as it will just pay those who already got tons of token for not transcoding but having a rpi staking.
@f1l1b0x So what happens in a few years time when the GPUs of the mining rigs start dying and mining only happens on ASICs? Livepeer dies as well? Or is the long term idea that Livepeer solely relies on gaming PCs that connect to a service live livepool?
I really don’t see why it shouldn’t be worth it to buy dedicated hardware for transcoding if power and bandwidth is already available. Especially since capable hardware gets cheaper and cheaper - e.g. the jetson nano for <$100.
Also it’s not about upfront payment, it’s about “if I had transcoding hardware, I would start earning some ETH”.

[16-May-20 06:15 AM] f1l1b0x#6391
“what is in a few years when”…

[16-May-20 06:15 AM] f1l1b0x#6391
I saw gpu mining die a couple of times now

[16-May-20 06:16 AM] f1l1b0x#6391
the answer is we will adapt

[16-May-20 06:23 AM] f1l1b0x#6391
I guess many Os that wont run own equipment will do a public pool and minimize their opex cost by doing that

[16-May-20 06:24 AM] vires-in-numeris#5324
Well I guess if your target market for transcoders is so specific, the current incentives are indeed wrong. But then the whole decentralization idea and the merkle mine is also highly questionable…

[16-May-20 06:24 AM] f1l1b0x#6391
others with a few mining rigs will be easily running the livepeer software next to the hashing without big extra cost except bandwidth

[16-May-20 06:24 AM] f1l1b0x#6391
why? cause the networks expects decentralized participators to work?

[16-May-20 06:25 AM] f1l1b0x#6391
I mean are you worried about new transcoders being disincentivized to join?

[16-May-20 06:27 AM] f1l1b0x#6391
the whole idea with the delegation is to enable those who can work and to have a disincentive against bad performance

[16-May-20 06:27 AM] f1l1b0x#6391
we will see this either work or fail very soon with more load on the network

[16-May-20 06:28 AM] vires-in-numeris#5324
And how does a new transcoder get delegation?

[16-May-20 06:28 AM] f1l1b0x#6391
the same way a existing transcoder does

[16-May-20 06:29 AM] vires-in-numeris#5324
Without LPT inflation, a new transcoder needs work to get delegation. And delegation to get work

[16-May-20 06:29 AM] f1l1b0x#6391
not necessarily

[16-May-20 06:30 AM] f1l1b0x#6391
if you cover a location in which other OTs with more stake have a bad connection to the B will filter them out and work with you even so you have very little stake

[16-May-20 06:30 AM] f1l1b0x#6391
if you show good process on the network delegators eventually will bond to you

[16-May-20 06:31 AM] f1l1b0x#6391
that makes it very practical for us to enforce a global footprint and allows competition on a location based level

[16-May-20 06:31 AM] f1l1b0x#6391
the problem is with too much inflation rewards the OT might not even care to perform bad

[16-May-20 06:33 AM] vires-in-numeris#5324
If fees start to roll in, delegators would automatically switch to the OTs that actually get some fees

[16-May-20 06:33 AM] f1l1b0x#6391
correct

[16-May-20 06:33 AM] f1l1b0x#6391
he would have to share the revenue tho

[16-May-20 06:35 AM] f1l1b0x#6391
so new transcoders with little stake have a higher interest to share their rev than old ones who already have the stake

[16-May-20 06:39 AM] vires-in-numeris#5324
But with only a little stake it might take a while until the fees start to add up - unless it’s really the first T in this location.
And during this time the T needs to share a high proportion of the revenue, so the question is if the T would actually stay and not say “well this is not worth my time”

[16-May-20 06:40 AM] Nelson#6080
how long do we believe orchestrators will continue operating at a loss when the inflation rate hits 0%?

[16-May-20 06:41 AM] Nelson#6080
or transcoders

[16-May-20 06:42 AM] Nelson#6080
even minor losses in opportunity cost vs running your mining operation at 100% efficiency will lead many to stop running infrastructure for the network

[16-May-20 06:42 AM] Nelson#6080
if we look at the token distribution, we see that the largest party impacted by the inflation is really just tokenholders that were airdropped token during the merkle mine and remain inactive

{Attachments}

[16-May-20 06:43 AM] Nelson#6080
while orchestrators may stand to benefit at minimal cost, is it really a problem if inactive participants are gradually inflated away over time?

[16-May-20 06:46 AM] Nelson#6080
while I agree we can’t subsidize the network forever, I’m not sure forcing orchestrators to operate at a loss for an extended period is in the networks best interest

[16-May-20 06:57 AM] vires-in-numeris#5324
Livepeer would end up with only a few orchestrators, making it really hard for new transcoders/orchestrators to join and get some stake. New transcoders would need to work with those few orchestrators who will dictate their terms.

[16-May-20 07:02 AM] f1l1b0x#6391
so many assumptions but I agree there is a chance on those capable of doing much work to get most of the work and new transcoders can only compete against it with similar opportunity (idle capacities on mining rigs)

[16-May-20 07:04 AM] f1l1b0x#6391
I am in favor of this idea since I dont want transcoding price to be artificially high so that transcoders can pay for their hardware but those with existing hw capacities should get the chance on earning from it.

[16-May-20 07:04 AM] f1l1b0x#6391
individuals with only few gpus will end up in public pools

[16-May-20 07:10 AM] vires-in-numeris#5324
@f1l1b0x any comment on this: "I really don’t see why it shouldn’t be worth it to buy dedicated hardware for transcoding if power and bandwidth is already available. Especially since capable hardware gets cheaper and cheaper - e.g. the jetson nano for <$100. "
Apart from not being currently supported…

[16-May-20 07:10 AM] f1l1b0x#6391
compare it to someone who has already a rig running

[16-May-20 07:18 AM] vires-in-numeris#5324
The jetson could do 4x/8x 1080p30, at 20k per pixel, that’s ~ 4x times $0.40 per hour (minus power, which is really low for the jetson).
I would be more than happy with $1 per day - the jetson will pay for itself in 4 months.

[16-May-20 07:21 AM] vires-in-numeris#5324
Would be interesting to see how low a mining rig could go price wise (calculating in the % loss in performance, the slightly higher power consumption and the bandwidth restrictions)

[16-May-20 07:23 AM] vires-in-numeris#5324
so theoretically, with 12 hours per day streaming 4 1080p streams I would be fine with a 1k wei pixel price (at $200 ETH) on the jetson

[16-May-20 09:32 AM] velvetdoctor#3682
Just watched the community call and went through the latest discussions. Super interesting!!!
Personally, I agree with Philippe and Doug that UNLESS necessary, it is better not to change the monetary policy and target inflation rate.
Even if the inflation goes to zero and some orchestrators exit, it will cause the staking percentage to fall and if the stake falls below 50%, inflation will however start again. If the stake doesn’t fall, then it means lot of them who are staking are hopeful that Livepeer is going to generate enough fees on network which would compensate them handsomely for staying back. Now how long they would wait depends on how quickly Livepeer start getting some meaningful demand. And we however need those long term believers in the network and not those short term Os or Ts who are here for just inflationary rewards.
Second thing, I heard people proposing a higher target stake %. I even heard something like 80-90% also on higher side. I think that is almost impossible to achieve. Why? Because (the newcomers may not be aware of this) in the first 10 million LPT that was released, 6.3 million went for airdrop and from which my estimation is that only half of it went to the miners and another half (approx 3 million) went to those addresses who might never care about the LPT in their ETH addresses and those tokens are effectively burned. That means almost 15% of current token supply (20mn), will never be bonded!! So even though we see 67% as current bonding rate of total supply of 20mn, I would say it is actually about 85% of stake. So if we increase staking target rate to 66% or higher then we will be left with only 20% LPT in circulation for liquidity which I don’t think is enough for secondary market or other Defi purposes that might come up in the future.
Anyway, I thought one should keep in mind that 15% of LPT supply is effectively lost and will never be bonded before increasing target bonding rate any higher. Personally I am happy with 50%.

[16-May-20 09:37 AM] f1l1b0x#6391
well the tokens are not burned of course but imobile

[16-May-20 09:37 AM] velvetdoctor#3682
But all these discussions are good and insightful and I am learning a lot. Thanks to @Nelson and others for bringing this up and having a lively discussion with the network best interest in mind.
I am pretty excited to see where it all goes in the next 6 -12 months. :blush:

[16-May-20 09:38 AM] f1l1b0x#6391
me too

[16-May-20 09:38 AM] f1l1b0x#6391
we are about to start marketing in the next few weeks

[16-May-20 09:38 AM] f1l1b0x#6391
we wouldnt want to burn our funds before we have a product that fits the market

[16-May-20 09:39 AM] velvetdoctor#3682
@f1l1b0x yeah may be “burned” is not the right word but kind of :blush:

[16-May-20 09:39 AM] f1l1b0x#6391
we are still a bit before product market fit but the next months and our first customers that use transcoding at scale will change that

{Reactions}
:pray: (2)

[16-May-20 10:46 AM] alvin#3301
I also agree not to change the current inflation policy. But I voted yes hahaha :joy::grin:

[17-May-20 10:08 AM] NicoV#7103
There is a fundamentally wrong view that the end of inflation before fees is a bad thing. It’s not, it’s the best thing that can happen to the protocol actually. It will root out orchestrators that can’t be competitive. Anyone claiming otherwise, is quite frankly just thinking about themselves and not what is best for the protocol in the long run.

As long as this inflation keeps up we will have the current status quo as a result whereby operators don’t want to pay upfront for hardware to run a proper node or be willing to invest time in contracting GPU miners and running a service that way. I see way too less of that happening while there has been plenty of incentive to do so. So if those operators can’t act altruistically when the livepeer protocol has given them so much in terms of rewards vs the actual resources needed to operate pre-streamflow then my opinion is to kindly point you to the exit.

The current inflation structure has created too much of a stigma around LPT which is “I need LPT to get inflation rewards” rather than “I need LPT to get more work and earn more ETH fees” which has a detrimental effect to the network if it keeps up. I have nothing against staking-as-a-service business but they are not adding anything of value right now and just diluting operators that do want to run capacity. Those node operators that are just there for LPT rewards need to be get rid of ASAP. It’s just pure kafka that you can be profitable while not supporting the network and it’s been my biggest point of critique on the protocol up until now.

[17-May-20 10:14 AM] NicoV#7103
I also feel like the current mechanism is being totally forgotten. Operators that don’t want to run capacity when inflation hits 0 will likely be unbonding and selling on the secondary market this will in turn drive the participation rate back down and could increase inflation again. This cycle is likely to continue until you end up with a set of operators that do want to seed capacity on the network once meaningful fees start flowing through the network.

Is the current setup ideal? No see my critique above but we have never even seen this part of the design play out in the wild. Let’s see if it works as intended first before wildly trying to rework the economics of the protocol.

Also what’s the chance we settle on a final design and are able to implement it and audit it within that 7 month timeframe given the highly contentious nature of this topic? It’s slim. Let’s see how the protocol plays out as it was originally intended and I think that will be just fine.

{Reactions}
:+1:

[17-May-20 10:15 AM] NicoV#7103
Anyway the gist of my rant is that prolonging LPT rewards isn’t going to change anything to the status quo of some operators not wanting to make time or hardware investment to add value to the network. That requires a more considered change in the economic curves.

[17-May-20 10:17 AM] NicoV#7103
so yeah I second what @f1l1b0x says entirely

[17-May-20 10:22 AM] NicoV#7103

I’m not sure forcing orchestrators to operate at a loss for an extended period is in the networks best interest

Please tell me how a 80$ VM a month while getting 1K$ USD in LPT value EACH DAY is operating at a loss?

[17-May-20 10:26 AM] NicoV#7103
Or here’s another example. 1.4M stake , no earned fees so not doing anything for the network. 36K earned LPT just through the reward cut of the orchestrator itself (not counting the actual amount made on its principal). Given a 2 year lifespan of the protocol so far that’s 18K LPT a year.

[17-May-20 10:51 AM] Nelson#6080
I’m not sure I agree, if there were steady fees flowing through the network, delegations would naturally begin flowing to orchestrators that are running hardware and performing real work. Unless fees pickup in a very meaningful way, we are expecting these operators to run at a loss once inflation hits 0%. Is a Livepeer with 2 orchestrators left in the best interest of the network?

[17-May-20 10:57 AM] Nelson#6080
I’m also not sure the participation rate would fall below 50%, the only use cases for LPT are staking, providing liquidity on Polo or Uniswap and perhaps lending in the future. If one holder of LPT sells LPT, they are most likely selling to another party that plans to stake LPT as that is the primary use case. The only way it falls below 50% participation is if 20%+ of the network are all unstaking to try and sell at the same time.

[17-May-20 11:03 AM] Nelson#6080
I agree many orchestrators are free riding here, but it’s mostly at the expense of airdropped token holders who are not actively participating in the network.

[17-May-20 11:03 AM] Nelson#6080
I’m not claiming to have all the answers, but I do think having constructive conversations with the community on these topics is helpful for everyone

[17-May-20 11:40 AM] f1l1b0x#6391
nevertheless we have to proof we can sustain the network with fees

[17-May-20 04:23 PM] NicoV#7103
I’d rather have 5 highly performant orchestrators of which the number grows over time as demand grows than 60 of which 55 are free riding on LPT rewards, yes. I think most would agree. The inflation has given enough opportunity especially to large stakers to build out infrastructure or value added services and partnerships over time which hasn’t happened so yes let us disrupt the status quo in 7 months. If it creates downward price pressure on the token so be it if that lowers the barrier of entry to operators that do want to run transcode capacity.

Which plays into the cards of the proven fact that in every industry whenever competitors fall out of the market they will be replaced when the opportunity arises.

And yes fees > inflation !

[17-May-20 04:26 PM] NicoV#7103
I’d also like to point out that the perspective of my opinion here is more as an orchestrator on the network that has taken considerable effort to build a value added service in my free time without hundreds of thousands of LPT backing me to do so , not as a core dev of livepeer.