45 comments

[ 3.7 ms ] story [ 163 ms ] thread
My personal opinion a central reason for the fading of volunteer-based distributed scientific computing (such as the mentioned SETI@home) is the rise of laptops.

Normal desktop PCs (in particular the ones that one builds by oneself for gaming) have sufficient cooling capacity that it is causes no serious issue to let them run to help some interesting science projects while you are at work. If you use a laptop, this is much more inconvenient since the cooling design of laptops is nearly always a compromise, and they are typically rather not built to run many hours of "numbercrunching" on most days for months.

EDIT: the article also at least as a sideline mentions this fact: "Making matters worse, fewer people have this hardware due to the general shift from desktop computers to laptops."

When I was a wee lad in my parents' basement with bread racks full of old x86 boxes cranking away on Distributed.net's [1] RSA cracking projects, distributed computing was in fashion because there was a lot of spare compute globally. As you mentioned, a lot of hardware has shifted to laptops. That is one part of the bifurcation. The other part is the mass movement to cloud infra, where there is much less idle compute available. Distributed computing was arbing excess idle compute at scale, which has seen a decline due to the factors mentioned.

The EFF DES cracker was also pretty cool in a hybrid distributed computing/custom ASIC fashion [2] [3] (which partnered with distributed.net to chew through key space for RSA Security's DES Challenge III after having accomplished a brute force challenge exclusively).

[1] https://en.wikipedia.org/wiki/Distributed.net

[2] https://en.wikipedia.org/wiki/EFF_DES_cracker

[3] https://archive.org/details/crackingdessecre00elec

> That is one part of the bifurcation. The other part is the mass movement to cloud infra, where there is much less "spare compute" available.

What was moved to cloud infrastructure are rather servers that were on-premise in the company's data center in the past. These on-premise servers did never really have any "spare capacity" that would have been allowed to be used for such scientific distrubuted computing projects.

There were some folks contributing to distributed.net that had entire corporate server fleets contributing. It was a different time. I don't know if the leaderboard is still available or archived, same with mailing list and IRC where people were admitting to it. Similarly, people in an IT admin capacity might push out a background process (distributed.net had a binary that would run as a service with no user interface exposed) to fleets of corp desktops with their identifier in the config for leaderboard purposes.

https://www.distributed.net/Policy ("distributed.net does not condone the unauthorized use of its software on any computer system. You may not run any distributed.net software on a system unless you own the system or have received permission from the owner to run distributed.net software. Running the client on a machine without authorization will result in your removal from the project and will disqualify you from winning.") [My note: This exists because of the above mentioned naughty behavior]

I should go digging for those EFnet IRC logs.

People got fired for running distributed.net on company/school infra more than once.
I think even with desktops is that they consume much less power, generate less heat and make less noise (fans run slower) when idle vs. operating at max power on distributed computing tasks. This wasn't historically the case.
Electricity prices, too. It'd cost me upwards of a 100 bucks per month to contribute to one of the @home projects. Not exactly pocket change.

It's also inherently inefficient, iirc every work package is dispatched to several clients and the majority result is used.

>There seems to be much less excitement about distributed computing these days.

Are they ignoring the hype around AI?

For which distributed computing project involving AI can one provide computing time?
That is not practical due to latency issues. A cluster of machines owned by one entity is still distributed computing.
The article is exactly about projects like SETI@home or Folding@home.
But those are just a subset of distributed computing which is itself a very large category of things. Even a single machine can be distributed computing the computation is distributed between multiple processes.
Yes, but the article is still about volunteer computing networks like SETI.

>Even a single machine can be distributed computing the computation is distributed between multiple processes.

No.

> But those are just a subset of distributed computing

Yes, but you must either have 1) not read even the first sentence of the article, or 2) be intentionally obtuse and obnoxious to not recognize that this article is only about "donate your spare computer time" type of distributed computing. Headline is somewhat unfortunate, but to pretend that you somehow didn't understand what the article is about is considerably more unfortunate.

The author doesn't understand what distributed computing means and so has decided to devote an article entirely to volunteer computing projects.
It's not dead. We just call it cloud computing now. After all what is a Kubernetes auto scaling solution if not distributed? Lambda/serverless compute even more so.

Also, in the heyday of Seti@home, computers were usually left on and didn't use a whole lot less power when idle because power management was far less advanced than it is now. Intel power states and gating were only just in their infancy. CPUs ran at fixed clock rates. And energy was cheap. So people didn't care.

These days it costs real money because idle time is much cheaper than heavy loaded time, and it's better spent financially donated to the organisations needing the compute and spent on cloud computing where they can manage the workload directly.

> It's not dead. We just call it cloud computing now.

Cloud computing is different: I cannot provide computation time of my idling computers to cloud computing (like I can for, say, SETI@home).

It's still distributed and still running on someone else's computer. It's kinda the same thing. Just the evolution of it.

And your idling computers won't be idling anymore when they're running SETI@home. These days there a huge energy difference between idle and load. It just doesn't make sense anymore for people to contribute that way. And for institutions that need compute to set up the whole infrastructure, software toolchain, enduser support etc around making it possible. Better to just pay Amazon.

To be more precise, it's queue based / batch processing; if it has to be done on the cheap, that's the spot market.
>Distributed computing erupted onto the scene in 1999 with the release of SETI@home, a nifty program and screensaver (back when people still used those) that sifted through radio telescope signals for signs of alien life.

Distributed computing erupted onto the scene in the early 80s when microprocessors became mainstream.

I remember running Prime95 in 1997. I guess SETI@HOME just got press?
I remember lots of my family running SETI@Home in the late 90s/early 00s, none of whom are in IT (but do tend to gravitate towards science/engineering). Also some Folding@Home at this time. I don't recall anyone running Prime95.

SETI@Home may not have been the first of its type, but it certainly seemed to have been the first that became broadly known (and together with Folding@HOME, arguably only one of two that ever did this).

The early version of SETI@Home you'd install the same as you installed a screensaver. It was trivially easy to get running, basically no setup.
This article suffers from a bad vocabulary collision. Distributed computing existed long before SETI@home, and will continue to exist as long as there isn’t just one computer doing everything.

“Volunteer computing” is probably a better way to refer to this phenomenon.

That’s a special case use of “distributed computing” unworthy of ars technica.
I didn't see any mention of cryptocurrency mining in the article.

At one point, people would throw their personal and employer's computers at one of the volunteer projects (SETI, Folding, various hash/key cracking) because that was a cool thing to do.

Some early Bitcoin mining was done for similar reasons.

Then cryptocurrency started to be worth money, and mining pools attracted a whole lot of other people to this 'distributed computing', and also attracted at least some of the people previously contributing to altruistic volunteer compute projects.

Somewhere in there, maybe because of mining awareness, it became less acceptable to run non-business things on employer's idle computers, so that's a lot more computers removed from the pool.

SETI@Home was an easy sell early on - "Hey, check out this neat screensaver" - but it got a lot harder to use when they joined the BOINC platform. That was when I abandoned it because I just plain had trouble getting it to run.
That's it, I'm sure there were some paid distributed computing projects that you could earn money for, but nothing on the scale of bitcoin and co. Seti and Folding were for bragging rights, bitcoin was for money, and I'm 99% confident that the people already doing those were some of the first to switch their rigs to crypto and were some that actually made money off of mining.
It is neither dying nor fading into the background. Huge numbers of scientific endeavours (eg LIGO) depend on distributed computing. Much of the world of finance depends on massive risk models computed using distributed computing. The LLMs driving the AI hype cycle are for sure not trained and finetuned on single machines etc.

A very specific old model of distributed computing isn't as popular as it once was because the world of technology has changed beyond all recognition since the likes of SETI@Home etc.

I think the term "distributed computing" used in the article has a different meaning then "distributed systems" which you are referring too.

Think of it as peer to peer compute. Like bit torrent but instead of data transfer, you compute things.

Crypto is the latest iteration of this. Certainly isn't main stream

I understand completely what the author wants distributed computing to mean. I'm also very familiar with the scene he's talking about. In my time I ran Seti@Home, Folding@Home<- (? I think that's what the thing was called), the great internet mersenne prime search etc. I've even done the crowd-sourced distributed computing thing in Eve Online.

However this kind of crowd-sourced computing is a tiny subset of distributed computing. Unless you think (for example) that the author has got it right, and me and Leslie Lamport have got the definition wrong[1].

[1] https://dl.acm.org/doi/abs/10.1145/3335772.3335934

> Folding@Home<- (? I think that's what the thing was called)

If only the article could help you out in finding its name.

Since it's got completely the wrong end of the stick about distributed computing I'm not sure it can be trusted on names of things in general.

That being said, I know the article mentions Folding@Home and the It Project. I know I ran a distributed protein folding thing back in about 2002. I'm not sure it was "Folding@Home" though. That's not something the article can help me with.

We are in agreement. The current state of the art iteration of that "scene" is now in cryptocurrencies.
The economy works because there's specialization and different products. Thus I can trade product A for product B and both me and the person I traded with actually gain something we didn't have before.

The problem with distributed computing is that there's no specialization. My compute is the same as your compute. There's no way where I can trade compute where we both benefit.

So in the end, even crypto won't really work either. You basically have to print tokens for crypto to work and that's essentially the same as giving you a piece of paper and calling it a dollar as payment.

Wasn't this article reviewed by anyone? Distributed computing is used everyday
you're thinking of industrial compute like a cloud

author and some people here (including me) correctly recognise the intended meaning of "compute gifted by casual people" - like SETI or RSA projects

The marginal cost of using spare capacity in home computers has risen significantly. It used to be the case that CPUs pretty much always used the same amount of power regardless of load, so filling those unused cycles with SETI@home et al. was essentially free. Now, whether the CPU is idling or running full bore makes a significant difference both in energy usage and in noise development. I would not want my desktop's fans to drone away while I'm trying to work.
A pretty large nit: "Twenty-one years later, SETI@home shut down, having found nothing. An incalculable amount of PC cycles and electricity wasted for nothing."

Negative results are still results. What SETI@Home showed was that after analysing all available data exhaustively that there was in fact no clear sign of alien intelligence to be found. That's a different state of knowledge than simply assuming the same result without examination of data.

It's a result that's disappointing to some, no doubt. It is still, however, a net increment in the amount of human knowledge.