back
80 comments
Oddly, they don't seem to support or want the providers to be running VMs. That moves it out of "idle hardware" into "dedicated hardware" territory right there. I can dedicate a VM as I can restrict that VMs resource limits, network access, and it's separate from my processes. But even if I had bare metal servers that weren't running VMware, Solaris, FreeBSD I'm certainly not going to let their provider service run on my hardware next to whatever else is on there that's mine.

My idle compute resources all run VMWare.

This is definitely my biggest issue at the moment. I understand the intent is to make it easy to port software to support it but I dislike the method none-the-less.

I'd much prefer they provide a constrained environment and dedicated compilers (or compiler extensions). At least that way the code can be sandboxed in such a way that you don't have to worry about what it's actually doing.

Honestly with the types of tasks they are targetting, a purely functional or algorithmic language requirement would serve them well. For example, I would be far more comfortable if they were using something like [Halide](https://halide-lang.org/) with some static analysis to deny any low level fuckery. At least that way the providers could have a reasonable assurance that they aren't going to accidentally unleash some malicious code onto their system.

They support WASM-based workloads that are inherently very sandboxed by design (and have other benefits like being architecture agnostic).
I think it's because some configurations of Golem use virtualization features to sandbox individual workloads, and some platforms have limitations around VMs-in-VMs.
This appears to have a bit in common with an idea I was toying with a month or two ago. I see nothing in this implementation that actually encourages decentralization, though. The economics of providing compute services certainly won't do it by themselves. We already have a networks of computers and ways to pay for them, and that has only encouraged centralization into a handful of cloud compute providers.

I can certainly see how something like this can reduce the barrier to entry for becoming a small-time compute provider, but I don't see how it changes the economics: specialization and centralization (up to a point) is cheaper.

Is there something I'm missing?

Well, as I think about it, it's interesting because although you're right that specialization and centralization is cheaper, if I'm Ordinary Joe and I want to make some money while my home PC would otherwise be idling, I can still sell my compute resources cheaper than someone running it as a serious business because anything is more than zero and I have essentially no upkeep costs to cover.

Granted, that kind of resource is only fit for some kinds of workloads, and I could be thinking about it wrong. But if this holds up, then I'd think the bottom's about to drop out of the cost of low-requirement cloud computing even though I expect this will not challenge the optimized cloud giants for large workloads.

Remember Filecoin, where you can trade storage in a similar way? Does anybody really store files in it?
You can see the live file storage deals here. https://filfox.info/en/deal
I had started reading about that and Siacoin a while back. Never got around to starting a node or anything like that. With Sia at least, you had to put up a stake first before you could start a storage node (for reasons that made sense). But then I'd have had to go through the rigamarole with buying some other coin, trading for it, setting up various wallets, etc.

Maybe at the end of the day I should just be giving a little money to rsync.net or Backblaze.

You can use Sia from browser nowadays. They built a decentralized CDN network called Skynet on it. https://siasky.net

You can host web apps on it too, here's the "Skynet app store": https://skapp.hns.siasky.net/#/apps/all

Yeah, it's so cheap to store on Filecoin I really want to backup to it as another offsite option. But I'm waiting until Duplicati has built in support for it instead of figuring it out on my own.
Good question. Would be nice if they would. Does anybody really uses Bitcoin?
The first question I have is whether a resource lender have a say in what's being computed on their resource?

If yes, it's not 'censorship-resistant'. If no, how can one be sure he's not transcoding CP at home?

This is a hard problem, but usually gets handwaved away to "that's the cost to play". Distributed storage has the same problem - how do you know you're not storing CP? - and there are some novel approaches to attempting to solve that problem (content hash checks against known material, IP list flagging, etc...) but at the end of the day you basically have to admit you can't solve that problem. That's the price of "censorship resistance".
Anonymity isn't the only possible solution to censorship resistance. Transparency + decentralization is another.
I think the more important question is:

Does the foreign code my machine runs have Internet access?

Or can it only send the work result to the Golem network itself?

I wouldn't want my IP to be used for DoS on the public Internet or whatever.

The answer is yes as long as you're using any browser, Golem wouldn't make it much worse.
Sounds like this could be summarized as a compute/currency exchange. Convert compute to currency which can be used to buy compute.

Does this have the built-in price correction of PoW crypto where economic viability determines how much compute is being dedicated to the network?

Aka, do prices naturally rise in this system until it's economically viable for someone to run their system?

>Does this have the built-in price correction of PoW crypto where economic viability determines how much compute is being dedicated to the network?

The answer is essentially "yes", but I think this question is showing a misunderstanding. Golem isn't the one creating the workloads that people are running, it's just connecting people selling compute power and people paying to run workloads. (The compute workloads aren't used as a consensus mechanism like PoW. Golem just relies on Ethereum to manage consensus.) The question is like "Does ebay implement a price correction system where economic viability determines how many resources people sell to it?".

Ebay actively manages supply through communicating with sellers, so "Yes?"
Similar posts should be clearly flagged in the title.

Even better if we could have user-configurable filters to hide unwanted topics.

Yes. There are alternative message boards that provide filtering (and many more useful features).

Hints: pincers, sideways walking, marine habitats.

I'd rather use https://lemmy.ml/ - it's implementing federation over ActivityPub.
I bought some GNT back around 2016-17 because I thought it was a cool concept. Held it for a few months, years maybe? I can't recall for certain. I do recall Brass Golem getting released which was supposed to be a big deal and I never saw anything change. So I sold my GNT and looking at the primer, I still don't see much of a change from a few years ago.

Mined some ETH at the same time and it is doing much better.

What is the greatest value/contribution in tools of this type?
you mean of these projects? Golem decentralizes computing power: censorship resistant, and resistant to central cloud failures (remember Google down in December)
I don't think i would ever want to be liable for the shenanigans of another programmer.

some of these applications will inevitably do some nasty things

It's cool to see that work has continued on this. I saw it posted a couple years ago and it seemed like just another shitcoin
This is kind of cool. It's like Urbit without... well, at least half the bullshit.
Is the admin password “emet”?
Is this another marketed solution that let's you do less than when not using it, effectiviley creating more problems without actually solving anything? Or am I just paranoid?
I mean it's basically just a zero trust fog compute platform. I apologise for the buzzwords but it's probably the most concise way to describe it.

On the Consumer/Requester side it let's you farm out a parallel and reproducible compute problem at whatever the going rate is based on market forces. If you can break out the heavy lifting in your code into parallel and deterministic/reproducible components then you can run compute fairly cheaply and quickly. For most people the zero-trust security isn't as important but the redundancy aspect of it is nice.

On the Producer/Compute side it lets you sell your unused processing power to do something actually useful. Think of it like BOINC but with a financial reward attached to it. While some people may eventually try to set up data centres as producers, for the most part the intent is to allow people to get some small profits back from lending out their compute power to scientific computations, 3d renders, simulations, etc.

Golem has been through a number of iterations over the past few years and is starting to come together in a fairly cohesive manner now. There are a lot of limitations at the moment but each time I've checked in on the project it's been a lot more versatile and a lot less restricted. I don't think it's ready for a lot of use cases yet however it is getting closer.

So I'd say it's worth playing around with if you like the tech however if you are expecting a well polished product I'd recommend keeping an eye on it for the next year or two.

I sort of despise the use of the word democratization to describe a system in which you pay for representation.
The submitted title was "Golem democratizes society’s access to computing power", which is a representative sentence from the article, but since it turned out to be baitier than the main title, we've replaced it with the main title above.
"Democratize" has joined "revolutionize" and "disrupt" on the list of buzzwords that make my eyes roll so far they glaze over. The words may be apt, but they need a fallow decade before being meaningful again.

Time to add "democratize" to mourner.github.io/bullshit.js. (I haven't seen it yet, but I'm predicting "catalyze" will join the list before the decade is out.)

Oligarchize doesn't quite have the same ring to it.
Democratize can also just mean “make more accessible to everyday people.” Which leads to the interesting and possible scenario that one can be against democracy (as a political system) but for democratization.

Apple (at least back in the 80s) could probably be described this way. Democratizing computer technology but operated as a secretive hierarchical organization.

It's an open marketplace for anyone to sell or buy compute power, not a giveaway from someone. I think the word "democratize" is clunky but I think it has some overlap with the idea that you don't need special connections to trade in the system and act as a seller or a buyer in it.
I rather like it, since it is in many ways an accurate description of how representation of how the democratic process works in practice.

Voters pick winners, but get little representation, people who fund campaigns of the winners get a lot of representation.

Yeah, it is also very similar how US has democratized middle east for the past 20 years
you mostly pay for it with taxes, or other people's labour (slaves)
Despise, sure, but accurate.
If you wouldn't pay your taxes, you wouldn't be able to vote. Paying is one of the main part of our society, and I wish I'd see less HN'ers obsessing over anything with a price.
Anything that uses the word democratized is something I'm not interested in. It's scary that paying money is considered to be some kind of representation.
The submitted title was "Golem democratizes society’s access to computing power", which is a representative sentence from the article, but since it turned out to be baitier than the main title, we've replaced it with the main title above.
Sounds very American.
That's not the meaning they are going for. It means allowing everyone to have a chance. Like when Unreal Engine or Unity said they democratized game development by releasing their engines for free.