this is a huge disrespect to the whole users, investors and your customers, “unstoppable “ the unstoppable network is down more often than solana at this point, what is the point of having 13 nodes if they act like a single server, if 1
Breaks all of them breaks.
@dominicwilliams says “ if you deploy software on ICP it will never go down” well, that’s not true, the NNS went down, today more subnets are down, if the answer is “it’s due to upgrades” do you think that’s an excuse? Wasn’t windows outage due to the same reason ? An upgrade? So how different is ICP compared to centralized servers running software if anyway after an upgrade the whole server is going down?
13 nodes and all of them act as a single server, where is the resilience in practice? Is just theory ? ICP slowly losing its value prop sadly.
It’s also concerning how we have “normalized” this kind of events, I remember when the NNS went down, we were crazy asking for clarification, now this events are usual and they don’t even provide any post mortem anymore.. is ICP losing its focus? Have we officially “relaxed” ?
As there is no leadership on top, this helps swamp to get control.
It looks to be just a maint or update. This is actually bad because subnet has to be up 24/7, even if 10 min down, it will push actual devs away. In future those interruptions should be less often and coordinated to fixed time.
Zforever97, on ICP, those 503 errors are usual thing during updates. I have not seen those few weeks now, so i think they appear less now.
I encountered a 400 error while transacting on ICPSwap, possibly due to subnet network delay. I hope DFINITY can address this soon to prevent further disruptions. If the infrastructure isn’t solid, flashy apps like CaffeineAI are meaningless.
We don’t have any reply from the team when they are being criticized (with facts, not fud) they seem to disappear when we are discussing this kind of things, YES DFINITY SUBNETS ARE DOWN, and you guys are taking millions a month from US, the investors, I don’t want to see a single subnet down from now, otherwise this weaknesses will start to be exposed.
I didn’t notice anything down but even when I have things have been restored in record time and DFINITY staffers have been very helpful and attentive even during weekends or off hours. You cannot compare a brief outage to Solana being down for days or even an entire week.
I love how people love to complain but hardly ever show their gratitude for when everything is working and working very well for the last 4 years since Genesis I might add.
And despite the drama regarding SNS projects or any obstacles in the road, do you really think any project of this breadth and ambitiousness will not have some issues along the way?
ICP is REAL and the problems it is trying to solve are REAL.
So if a datacenter is down on Amazon, is it fair to expect Jeff Bezos to respond directly to you?
LOL. I can see your hidden post and I get what you’re saying.
Thing is that the ultimate goal of the internet computer is to be a decentralized world computer that will have distributed your app across many geographic areas and subnets eventually.
As the network is still very young, it has not reached this goal just yet but the fact is that it is distributed to some degree and therefore there is a lot of built in redundancy and fault tolerance when you deploy on the internet computer not to mention load balancing as well I’m assuming.
It is a really great goal to eventually have “unstoppable” software but you’re right. At the moment, realistically, it is not truly unstoppable just yet however the dream is to have it built out enough and have it mature enough where this will truly be the case.
Just like how people are annoyed that DFINITY has 25% voting power right now. It was down to 19% when there were more investors and now has gone back up. The whole idea is to get enough participants where DFINITY’s voting power will eventually get below a certain amount. The network is too young and there is not enough adoption.
Adoption is the key here. You adopt technology in early stages and when more and more people adopt it, it gets better and better and better. The current internet was adopted commercially starting around 1995ish and look where it is now 30 years later. I remember how we still had dial up modems and horrible looking websites and the biggest thing back then was FLASH and you could barely do much other than send emails and see very rudimentary websites.
This is a very ambitious project and DFINTY is making great progress towards the vision imo. What we need are people to have a long term outlook here because DFINITY definitely knows it will take time and has a 20+ year roadmap for a reason.
The subnets are not going down but are being upgraded regularly, which I consider a superpower of ICP. Currently, an upgrade takes 3–5 minutes depending on the size of the subnet’s state (it used to be shorter when the state was smaller). Of course, I’m not at all happy about it taking this long, as canisters are unavailable during this period. In theory, downtime for query calls can be zero—since different replicas can upgrade at different times—but downtime for update calls cannot be avoided, as blocks can only be produced once a sufficient number of replicas have switched to the new version. However, this downtime can potentially be reduced to just a few seconds.
We are constantly working on improving ICP, and this is clearly a very important area. With the current approach, we expect to reduce update call downtime to 1–2 minutes soon. Bringing it down to just seconds will require substantial work and resources, which we unfortunately do not currently have—but it remains one of our key ambitions.
Of course, one could consider updating less frequently, but that also comes with trade-offs.
Thanks for the technical clarification Jan and nice to you see posting again. Has it really been 5 months? You’ve obviously been extremely busy! The fact that you believe that subnet upgrades could eventually just take seconds in the future with more work and resources is probably the most important key point. I have no doubt that this will be the case eventually. I feel that people are extremely unrealistic wanting Windows 12 to come out right after Windows 95 here if you get what I’m saying.
Could you possibly give us a clue about AI GPU acceleration though? Will GPU’s be deployed on a single subnet or multiple subnets in the future or will there be something like a Cerebras type of integration instead in order to speed up future AI deployments?
I realize that training is of course off chain but will Caffeine AI run completely as an on-chain dapp and will this tie up node machines if there are no gpus in the mix to begin with and will the network be able to support a large number of users in this case without crashing and burning? If running directly on the nodes, will it be fast enough for the current generation mix of node hardware?
Also, what is the status of WASM64 ? Is this about to be deployed into production soon? This is a required prerequisite for AI deployments needing a larger address space correct?
Didn’t realize. For some reason Jerry B. stated it wasn’t even released in one of his latest videos and I was scratching my head. I figured WASM64 was a requirement to have for AI in general. Thanks for the clarification Jan.
