Best settings for torrents #2328
-
Hello, I have a low-end machine seeding a few torrents stored in an S3 compatible storage (scw) using S3QL. I'm trying to move to JuiceFS but so far I'm getting unusable download speeds compared to S3QL (~12 MiB -> ~20 KiB). There has to be some overhead because the storage doesn't support random writes but not like this. I'm either hitting a bug or there's some issue with my setup. I have experimented with pretty much every option I could find but no luck.
The block size is 4 MiB. There also seems to be a lot of activity for a 200 MiB torrent. When I'm writing sequentially (e.g. v1.0.0-rc2 |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 5 replies
-
Hi, have you tried use |
Beta Was this translation helpful? Give feedback.
-
I started some benchmarks with garage/lmdb + juicefs/redis + qBittorrent here. Unfortunately the results are really bad. I'm not sure exactly why. Some people on torrent forums suggested that S3 is ill-suited for random writes and it would be much better to first download the torrent then "import" it to S3 for seeding. I'm sure there's a lot of tuning that can help achieve better performance and disk usage but i don't have energy to investigate (i've never used juiceFS before). You can see my findings so far on the link i just posted. Hopefully someone who is more familiar with juiceFS tuning can find reasonable settings to make it better. |
Beta Was this translation helpful? Give feedback.
-
I'm hosting my 'Home lab' setup on a stretched docker swarm. 1 X86 VM on Proxmox home. I'm running docker volumes and App / Media data directly from Wasabi storage, using VPS datacenter for speedy downloads. Torrents are slower, but that's more the nature of it (low reputation, not enough seeds etc etc), but NZB is quick! ![]() JuiceFS![]() TailscaleTailscale is connecting the node's, access only via Tailscale SSH, firewall is closed, apps open port(s) themselves from the inside, Cloudflare tunnels for remote access.
Drawing |
Beta Was this translation helpful? Give feedback.
Hi, have you tried use
juicefs profile
command to analyze the performance problem? And it seems like the latency of metadata engine is too high (50-100ms), maybe you should try Redis.