• Login
  • Register
  • Login Register
    Login
    Username/Email:
    Password:
    Or login with a social network below
  • Forum
  • Website
  • GitHub
  • Status
  • Translation
  • Features
  • Team
  • Rules
  • Help
  • Feeds
User Links
  • Login
  • Register
  • Login Register
    Login
    Username/Email:
    Password:
    Or login with a social network below

    Useful Links Forum Website GitHub Status Translation Features Team Rules Help Feeds
    Jellyfin Forum Support Troubleshooting Buffering when playing HEVC10 from SMB/NFS share

     
    • 0 Vote(s) - 0 Average

    Buffering when playing HEVC10 from SMB/NFS share

    shlajin
    Offline

    Junior Member

    Posts: 2
    Threads: 1
    Joined: 2024 Jul
    Reputation: 0
    #1
    2024-07-21, 09:14 PM
    Hello,

    I observe a strange behaviour when playing huge files (~90GB, HEVC10). I've found somewhat similara threads, but they all exhibit slightly different symptoms, so I figured I'd better start a new one.

    My setup:
    Synology NAS connected via a switch to a N100 mini computer. All have 1GPBs ports. N100 computer runs proxmox, and I tried running jellyfin as an LXC container, and as a docker container inside a debian VM. It gave me the very same results.

    When I play a big HEVC10 file (~90GB size) with bitrate of 120Mbps, the video is buffering for a couple of seconds ~5 seconds and I don't understand why.

    The reason I can think of could be the following:

    1) Transcoding: unlikely, as the playback info reports "Play method = direct streaming".
    2) Client bound: unlikely, as my current client is MBP with M1Max, should be plenty of compute power to handle the video
    3) HDD/SDD Bound: unlikely, as both mini computer and NAS should easily output more than 120Mb/s speeds.
    4) Network Bound: possibly, and that's where things are being weird.

    The NAS and mini computer are connected through a 1Gbps switch. I ran iperf3 on NAS and tested the speed from mini computer to the nas – the speed is stable 1Gbps. I did the same thing testing the speed from my macbook to the mini computer, and the speed is constantly about 350Mbps (stable 5Ghz wifi).

    Then I ran iftop inside the jellyfin container to see the network speeds. To my surprise, I observed that when I start playing the video, container started pulling data from NAS with the speeds about ~900Mb/s, and transmitting the data to my laptop with the speed of ~50Mbs/s. It clicked in my head – if my switch is only 1Gbps, it's just not enough bandwidth to receive and transmit! So I went to NAS and limited the NFS speeds to just about ~300Mb/s. However, it didn't change anything – the NFS speed was 300Mbs, and the streaming speed remained ~50Mbs.
    I decided to check if the NFS is the problem, and switch to SMB. Same thing, ~900Mb/s pulling from NAS as soon as the video starts, and the streaming remained as ~50Mbs. Then I decided to wait a little while – what happens after the ~90GB mark (i.e. when the whole movie is downloaded). And... nothing has happened – once it pulls ~90Gbs, it stops pulling from NFS/SMB and simply continues to play the video with the transmission speed of ~50Mbs, causing buffering every ~5s or so.

    As the final test, I benchmarked the SSD speed inside the container (well, who knows). Write speed is 1.1GBps (so 8+Gbps), read speed is 3.3GBps (so 24+Gbps).

    How can I fix this issue? Is it normal that jellyfin pulls the whole video via NFS/SMB as fast as it can?
    Thank you
    Fate
    Offline

    Member

    Posts: 219
    Threads: 3
    Joined: 2023 Jun
    Reputation: 16
    #2
    2024-07-21, 09:45 PM
    Your Apple m1 is probably a bad idea to test on since the CPU is not well supported since they are so rare. And I wouldn't call it fast as well.
    shlajin
    Offline

    Junior Member

    Posts: 2
    Threads: 1
    Joined: 2024 Jul
    Reputation: 0
    #3
    2024-07-22, 03:01 AM
    While it's certainly not the gaming-rig-level processor, it's hard to call it slow. Speaking with numbers, my 2 efficiency cores was under 50% each during the test, and the performance cores were less than 5% of utilization – I would expect at least one core to hit 90%+ before considering the client processor as the bottleneck. Obviously, thermal throttling also was out of equation due to temps being lower than 50C all the time. Also surprised to hear that M-processors are rare, given they are in each and every new mac for few years now.

    However, even though I'm skeptical of that's being the case, your point still stays valid. I haven't tested it on other hardware. I have a somewhat powerful gaming laptop (with 3070), will try a test on it soon. Thanks for the tip!
    « Next Oldest | Next Newest »

    Users browsing this thread: 1 Guest(s)


    • View a Printable Version
    • Subscribe to this thread
    Forum Jump:

    Home · Team · Help · Contact
    © Designed by D&D - Powered by MyBB
    L


    Jellyfin

    The Free Software Media System

    Linear Mode
    Threaded Mode