4 ms·

In my experience IPFS is only fast when you're using it through something like Cloudflare's IPFS proxy (which is basically a caching proxy). I haven't found IPF

In my experience IPFS is only fast when you're using it through something like Cloudflare's IPFS proxy (which is basically a caching proxy). I haven't found IPFS to be actually usable any time I've tried running a node myself. Is that unusual?

segfaultbuserr 6y ago
Interesting, didn't notice that before. I believed my downloads are too obscure to be cached by Cloudflare, but I may be wrong. I'll do a comparison once I fixed my server at home...
the_duke 6y ago
No. The last time I tried it (maybe 9-12 months ago) it was a real resource hog (including choking the network) and really slow.
magila 6y ago
This has been the state of IPFS for years now. The protocol and client haven't been developed to where they can work reliably in a real P2P scenario. IPFS only appears to work because most requests go strait to one of the big gateways which already has the content cached. Development of IPFS slowed to a crawl when most of Protocol Labs switched to working on Filecoin. It remains to be seen if either Filecoin or IPFS will ever be viable for their intended purpose.
- georgyo 6y ago
  I'm not sure you are right about this. The following link will stream big buck bunny directly from the network with full P2P support. It is not using a gateway node for the video content. https://bbb.fu.io/ https://bbb.fu.io/ You can run the following in the console to see all your peers for await (const peer of await node.swarm.peers()) { console.log(peer.addr.toString()) } The console currently get's spammed pretty hard because they haven't implemented filtering out plain ws (non-wss) peers if you are connecting https. But it will use webrtc to find peers, and find all wss nodes in the network.
  - magila 6y ago
    For me at least the vast majority of the data is being downloaded from various servers under bootstrap.libp2p.io or preload.ipfs.io. These are clearly dedicated servers not true peers. The contact info for these servers appears to be retrieved via the DNS rather than the DHT. This isn't surprising because browser clients can't participate in the DHT, but the DHT is currently one of the biggest weak points of IPFS. IPFS's DHT is so chatty that it consumes over a megabit of throughput just to maintain an idle node. Lookups are excruciatingly slow and often fail completely.
    - georgyo 6y ago
      Sorta, a peer can advertise dns addresses, and the browser can participate in the dht as a client. The example above uses webrtc-star for browser to browser discovery. $ ipfs dht findpeer 12D3KooWJxNHY6zE1KnzFdBJrTMCbhCVNtSLvjp7qR5Wsp49DnUC /dnsaddr/ipfs.scalable.io /dns4/ipfs.scalable.io/udp/4001/quic /dns6/ipfs.scalable.io/tcp/443/wss /dns6/ipfs.scalable.io/udp/4001/quic /dns4/ipfs.scalable.io/tcp/443/wss
  - haywirez 6y ago
    Took a good 10 seconds until the video started playing. Anything over a second is too long in terms of UX expectations.
- derefr 6y ago
  IPFS is the engine of Filecoin, though. Filecoin just incentivizes IPFS nodes to store things. So further IPFS development should still be necessary from the perspective of making Filecoin development/adoption happen. (I say this, and yet I know it isn't true in practice. But why isn't it true?)
  - nwah1 6y ago
    Pump up Filecoin price. Sell Filecoin. Move to Fiji. Don't do any coding while in Fiji
    - perlwle 6y ago
      Ha. I was super surprised that my dad(almost 70) mentioned IPFS to me the other day. It turns out the people he know are running or thinking of running IPFS nodes because of filecoin incentive. Apparently, the people don't want to miss another bitcoin opportunity. filecoin is also promoting in China. https://mp.weixin.qq.com/s/T0Qt2CEP7QsA6Zy9Vzvdkw https://mp.weixin.qq.com/s/T0Qt2CEP7QsA6Zy9Vzvdkw
- dsun179 6y ago
  Yes, this is still the case. For a file which is on exactly one other computer, you have about 45 seconds lookup time before the download can start.
swills 6y ago
Are you opening the required ports (4001, 8080)? I found it was like that too until I did opened the ports.
StavrosK 6y ago
I have to agree. I've had cases where I had to pin data that was on the node already and IPFS just hung the pin for minutes, failed, and then the GC removed the data later on. This is just one of the issues I had, other things include IPFS not being able to discover content on nodes that were already connected to it, IPFS just failing all my queued pins at once for some reason, etc. I regret ever building a pinning platform on IPFS. The worst thing is that, when I expressed my frustration on Twitter about an IPFS release completely breaking IPNS updates, the founder called me a Karen.

segfaultbuserr 6y ago

Interesting, didn't notice that before. I believed my downloads are too obscure to be cached by Cloudflare, but I may be wrong. I'll do a comparison once I fixed my server at home...

the_duke 6y ago

No. The last time I tried it (maybe 9-12 months ago) it was a real resource hog (including choking the network) and really slow.

magila 6y ago

This has been the state of IPFS for years now. The protocol and client haven't been developed to where they can work reliably in a real P2P scenario. IPFS only appears to work because most requests go strait to one of the big gateways which already has the content cached. Development of IPFS slowed to a crawl when most of Protocol Labs switched to working on Filecoin. It remains to be seen if either Filecoin or IPFS will ever be viable for their intended purpose.

georgyo 6y ago
I'm not sure you are right about this. The following link will stream big buck bunny directly from the network with full P2P support. It is not using a gateway node for the video content. https://bbb.fu.io/ https://bbb.fu.io/ You can run the following in the console to see all your peers for await (const peer of await node.swarm.peers()) { console.log(peer.addr.toString()) } The console currently get's spammed pretty hard because they haven't implemented filtering out plain ws (non-wss) peers if you are connecting https. But it will use webrtc to find peers, and find all wss nodes in the network.
- magila 6y ago
  For me at least the vast majority of the data is being downloaded from various servers under bootstrap.libp2p.io or preload.ipfs.io. These are clearly dedicated servers not true peers. The contact info for these servers appears to be retrieved via the DNS rather than the DHT. This isn't surprising because browser clients can't participate in the DHT, but the DHT is currently one of the biggest weak points of IPFS. IPFS's DHT is so chatty that it consumes over a megabit of throughput just to maintain an idle node. Lookups are excruciatingly slow and often fail completely.
  - georgyo 6y ago
    Sorta, a peer can advertise dns addresses, and the browser can participate in the dht as a client. The example above uses webrtc-star for browser to browser discovery. $ ipfs dht findpeer 12D3KooWJxNHY6zE1KnzFdBJrTMCbhCVNtSLvjp7qR5Wsp49DnUC /dnsaddr/ipfs.scalable.io /dns4/ipfs.scalable.io/udp/4001/quic /dns6/ipfs.scalable.io/tcp/443/wss /dns6/ipfs.scalable.io/udp/4001/quic /dns4/ipfs.scalable.io/tcp/443/wss
- haywirez 6y ago
  Took a good 10 seconds until the video started playing. Anything over a second is too long in terms of UX expectations.
derefr 6y ago
IPFS is the engine of Filecoin, though. Filecoin just incentivizes IPFS nodes to store things. So further IPFS development should still be necessary from the perspective of making Filecoin development/adoption happen. (I say this, and yet I know it isn't true in practice. But why isn't it true?)
- nwah1 6y ago
  Pump up Filecoin price. Sell Filecoin. Move to Fiji. Don't do any coding while in Fiji
  - perlwle 6y ago
    Ha. I was super surprised that my dad(almost 70) mentioned IPFS to me the other day. It turns out the people he know are running or thinking of running IPFS nodes because of filecoin incentive. Apparently, the people don't want to miss another bitcoin opportunity. filecoin is also promoting in China. https://mp.weixin.qq.com/s/T0Qt2CEP7QsA6Zy9Vzvdkw https://mp.weixin.qq.com/s/T0Qt2CEP7QsA6Zy9Vzvdkw
dsun179 6y ago
Yes, this is still the case. For a file which is on exactly one other computer, you have about 45 seconds lookup time before the download can start.

swills 6y ago

Are you opening the required ports (4001, 8080)? I found it was like that too until I did opened the ports.

StavrosK 6y ago

I have to agree. I've had cases where I had to pin data that was on the node already and IPFS just hung the pin for minutes, failed, and then the GC removed the data later on. This is just one of the issues I had, other things include IPFS not being able to discover content on nodes that were already connected to it, IPFS just failing all my queued pins at once for some reason, etc. I regret ever building a pinning platform on IPFS. The worst thing is that, when I expressed my frustration on Twitter about an IPFS release completely breaking IPNS updates, the founder called me a Karen.