!zghijEASpYQWYFzriI:nixos.org

Hydra

374 Members
109 Servers

Load older messages


SenderMessageTime
4 Apr 2024
@darkwater4213:matrix.orgdarkwater4213
In reply to @k900:0upti.me
No central server will not happen
Why not? What does the central server provide that can't be decentralized?
09:03:42
@k900:0upti.meK900 It could be decentralized 09:03:57
@darkwater4213:matrix.orgdarkwater4213
In reply to @k900:0upti.me
And realistically it's just not a thing most people will want or need
It'd be a proof-of-concept, to be sure.
09:03:58
@k900:0upti.meK900It just won't be Hydra anymore09:04:02
@k900:0upti.meK900It'll be some other thing you make09:04:05
@k900:0upti.meK900And it'll be easier for you to make it from scratch09:04:10
@k900:0upti.meK900Than it would be to reuse any of Hydra09:04:14
@darkwater4213:matrix.orgdarkwater4213One could argue that it would be more aptly named Hydra, for obvious reasons.09:04:27
@k900:0upti.meK900Because Hydra was never designed for this kind of setup09:04:27
@k900:0upti.meK900I don't care about the philosophical implications of naming things here09:04:56
@k900:0upti.meK900We already have a thing called Hydra and that thing can't do what you want, and probably won't ever be able to09:05:10
@k900:0upti.meK900If you make a new thing, you're free to call it whatever you want, including Hydra09:05:20
@k900:0upti.meK900(though it would be confusing for everyone involved)09:05:26
@darkwater4213:matrix.orgdarkwater4213

It's still a fascinating prospect.

I've been toying with the idea of NixOS-as-MDM for some time now. It just occurred to me as I was brushing my teeth that most MDM solutions require a central server (if not all) and that can get expensive and inconvenient. Decentralizing the entire thing makes it more portable, flexible, and frankly, way cooler.

I was thinking of a guy who installed some distributed-computation software on all his school's computers, so he could build an entire Gentoo system, DE and all, in ninety seconds flat (not counting transfer times). Then it occurred to me that if all those machines were the same architecture, there's no reason those compiled objects could not be reused with the exact same optimizations. And since there's already no "central server" (because all of them are servers), you could unify this entire mess into a NixOS config, and you've got yourself a hype train straight out of 2014.

I need sleep.

09:10:17
@k900:0upti.meK900MDM has absolutely nothing to do with Hydra09:10:48
@k900:0upti.meK900And honestly you're just describing distcc09:11:06
@k900:0upti.meK900(though it's still not entirely Decentralized(tm), not that it needs to be)09:11:26
@delroth:delroth.net@delroth:delroth.netany time someone suggests using IPFS I just assume they have no clue about anything related to distributed computing, so far the heuristic has always worked09:12:41
@k900:0upti.meK900Also, MDM and Decentralized(tm) don't go together09:12:45
@k900:0upti.meK900For reasons that should hopefully be obvious09:12:59
@rick:matrix.ciphernetics.nlRick (Mindavi)
In reply to @k900:0upti.me
Without effectively rewriting Hydra entirely
And nix too, since eval is quite memory hungry. Distribution of that will add a lot of transfer overhead and complication to evaluation, I feel.
12:11:32
@janne.hess:helsinki-systems.dedas_jRedacted or Malformed Event13:18:03
@mrtrk:matrix.orgMarco TurchettoHas anyone ever tried to deploy all the hydra components separately? from what I understood hydra architecture is meant to be distributed and use postgresql as a message bus13:48:12
@janne.hess:helsinki-systems.dedas_j
In reply to @mrtrk:matrix.org
Has anyone ever tried to deploy all the hydra components separately? from what I understood hydra architecture is meant to be distributed and use postgresql as a message bus
That's mostly true but I think (although I am not sure) that the queue runner expects the drvs from the evaluator in its store
13:51:47
@ma27:nicht-so.sexyma27
In reply to @janne.hess:helsinki-systems.de
That's mostly true but I think (although I am not sure) that the queue runner expects the drvs from the evaluator in its store
That should be solvable by instructing the evaluator to write into a remote store I guess.
13:53:45
9 Apr 2024
@5m5z3q888q5prxkg:chat.lightnovel-dungeon.de@5m5z3q888q5prxkg:chat.lightnovel-dungeon.de changed their profile picture.23:12:30
11 Apr 2024
@anthonyrsl:matrix.orgAnthony Rsl set a profile picture.21:59:20
@anthonyrsl:matrix.orgAnthony Rsl removed their profile picture.22:12:55
12 Apr 2024
@hexa:lossy.networkhexa joined the room.16:46:26
14 Apr 2024
@bootstrapper:matrix.org@bootstrapper:matrix.org changed their profile picture.04:54:39

Show newer messages


Back to Room ListRoom Version: 6