In a few months, I will have the space and infrastructure to join the selfhost community. I’m trying to prepare, as I know it can be challenging, but I somehow ended up with more questions than answers.

For context, I want to run a server with torrents, media (plex, Jellyfin or something else entirely - I didn’t make a decision yet), photos(Emmich, if its stable, or something else), Rook, Paperless, Home Assistant, Frigate, Adguard Home… Possibly lots more. Also, I will need storage - I’m planning for 3x18tb drives to begin with, but will certainly be adding more later.

My initial intention was to set up a NAS in Silverstone CS382(or Jonsbo N3/N5, if they’re in a reasonable price). I heard good things about Unraid and it’s capabilities of running docker. On the other hand, I’m hearing hood things about Proxmox or NixOS with NAS software running in a VM, too - but for Unraid, it seems hacky. Maybe I should run NAS and a separate server? That’d be more costly and seems like more work on maintenance with no real benefit. Maybe I should go with TrueNAS in a VM? If I don’t do anything other than NAS, TrueNAS shouldn’t be that hard to set up, right?

I’m also wondering whether I should go with Intel for QuickSync, AMD and Arc graphics or something else entirely. I’ve read that AV1 is getting popular, is AMD getting more support there? I will buy Intel if it’s clearly the better option, but I’m team Red and would prefer AMD.

Also, could anyone with a non-technical SO tell me how do they find your selhosted things? I’ve read about Cloudflare Tunnels and Tailscale, which will be a breeze for me, but I gotta think about other users aswell.

That’s another concern for me - am I correct in thinking Tailscale and Cloudflare Tunnels are all I need to access the server remotely? I will probably set up a PiKVM or the Risc one aswell, can it be exposed aswell? I will have a dream machine from Ubiqiti, anything that needs to run to access the server I may run there. I’m not looking to set up anything more complicated like Wireguard - it’s too much.

For additional context, I’m a software developer, I know my way with Docker and the command line and I consider myself to be tech savvy, but I’m not looking to spend every weekend reading changelogs and doing manual updates. I want to have an upgrade path (that’s why Im not going with Synology for example), but I also don’t want to obsess over it. Money isn’t much of an issue, I can spare 1-2k$ on the build, not including the drives.

Any feedback and suggestions appreciated :)

  • NuXCOM_90Percent
    link
    fedilink
    English
    arrow-up
    5
    ·
    edit-2
    14 days ago

    A LOT of questions there.

    Unraid vs Truenas vs Proxmox+Ceph vs Proxmox+ZFS for NAS: I am not sure if Unraid is ONLY a subscription these days (I think it was going that way?) but for a single machine NAS with a hodgepodge of drives, it is pretty much unbeatable.

    That said, it sounds like you are buying dedicated drives. There are a lot of arguments for not having large spinning disk drives (I think general wisdom is 12 TB is the biggest you should go for speed reasons?), but at 3x18 you aren’t going to really be upgrading any time soon. So Truenas or just a ZFS pool in Proxmox seems reasonable. Although, with only three drives you are in a weird spot regarding “raid” options. Seeing as I am already going to antagonize enough people by having an opinion, I’ll let someone else wage the holy war of RAID levels.

    I personally run Proxmox+Ceph across three machines (with one specifically set up to use Proxmox+ZFS+Ceph so I can take my essential data with me in an evacuation). It is overkill and Proxmox+ZFS is probably sufficient for your needs. The main difference is that your “NAS” is actually a mount that you expose via SMB and something like Cockpit. Apalrd did a REALLY good video on this that goes step by step and explains everything and it is well worth checking out https://www.youtube.com/watch?v=Hu3t8pcq8O0.

    Ceph is always the wrong decision. It is too slow for enterprise and too finicky for home use. That said, I use ceph and love it. Proxmox abstracts away most of the chaos but you still need to understand enough to set up pools and cephfs (at which point it is exactly like the zfs examples above). And I love that I can set redundancy settings for different pools (folders) of data. So my blu ray rips are pretty much YOLO with minimal redundancy. My personal documents have multiple full backups (and then get backed up to a different storage setup entirely). Just understand that you really need at least three nodes (“servers”) for that to make sense. But also? If you are expanding it is very possible to set up the ceph in parallel to your initial ZFS pool (using separate drives/OSDs), copy stuff over, and then cannibalize the old OSDs. Just understand that makes that initial upgrade more expensive because you need to be able to duplicate all of the data you care about.

    I know some people want really fancy NASes with twenty million access methods. I want an SMB share that I can see when I am on my local network. So… barebones cockpit exposing an SMB share is nice. And I have syncthing set up to access the same share for the purpose of saves for video games and so forth.

    Unraid vs Truenas vs Proxmox for Services: Personally? I prefer to just use Proxmox to set up a crapton of containers/vms. I used Unraid for years but the vast majority of tutorials and wisdom out there are just setting things up via something closer to proxmox. And it is often a struggle to replicate that in the Unraid gui (although I think level1techs have good resources on how to access the real interface which is REALLY good?).

    And my general experience is that truenas is mostly a worst of all worlds in every aspect and is really just there if you want something but are afraid of/smart enough not to use proxmox like a sicko.

    Processor and Graphics: it really depends on what you are doing. For what you listed? Only frigate will really take advantage and I just bought a Coral accelerator which is a lot cheaper than a GPU and tends to outperform them for the kind of inference that Frigate does. There is an argument for having a proper GPU for transcoding in Plex but… I’ve never seen a point in that.

    That said: A buddy of mine does the whole vlogger thing and some day soon we are going to set up a contract for me to sit down and set her up an exporting box (with likely use as a streaming box). But I need to do more research on what she actually needs and how best to handle that and she needs to figure out her budget for both materials and my time (the latter likely just being another case where she pays for my vacation and I am her camera guy for like half of it). But we probably will grab a cheap intel gpu for that.

    External access: Don’t do it, that is a great way to get hacked.

    That out of the way. My nextcloud is exposed to the outside world via a cloudflare tunnel. It fills me with anxiety but as long as you regularly update everything it is “fine”.

    My plex? I have a lifetime plex pass so I just use their services to access it remotely. And I think I pay an annual fee for homeassistant because I genuinely want to support that project.

    Everything else? I used to use wireguard (and openvpn before it) but actually switched to tailscale. I like the control that the former provided but much prefer the model where I expose individual services (well, VMs). Because it is nice to have access to my cockpit share when I want to grab a file in a hotel room. There is zero reason that anything needs access to my qbitorrent or calibre or opnsense setup. Let alone even seeing my desktop that I totally forgot to turn off.

    But the general idea I use for all my selfhosted services is: The vast majority of interactions should happen when I am at home on my home network. It is a special case if I ever need to access anything remotely and that is where tailscale comes in.

    Theoretically you can also do the same via wireguard and subnetting and vlans but I always found that to be a mess to provide access both locally and remotely and the end result is I get lazy. Also, Tailscale is just an app on basically any machine whereas wireguard tends to involve some commands or weird phone interactions.

    • ahal@lemmy.ca
      link
      fedilink
      English
      arrow-up
      1
      ·
      14 days ago

      You can still buy a lifetime subscription for Unraid, it’s just a lot more expensive.

    • sodamnfrolic@lemmy.sdf.orgOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      14 days ago

      I only said 3 drives because supposedly it’s better than 2, which would mean simple mirroring - I’ll cross that bridge when I get to it, shouldn’t be hard to get answers there.

      Is Cloudflare Tunnels really this problematic? I thought Tunnels and Tailscale would be safe… If I can’t expose those services, I’d rather pay for SaaS alternatives.

      Thank you for the other tips.

      • NuXCOM_90Percent
        link
        fedilink
        English
        arrow-up
        2
        arrow-down
        1
        ·
        edit-2
        14 days ago

        More drives is always better. But you need to understand how you are making it better.

        https://en.wikipedia.org/wiki/Standard_RAID_levels is a good breakdown of the different RAID levels. Those are slightly different depending on if you are doing “real”/hardware RAID or software raid (e.g. ZFS) but the principle holds true and the rest is just googling the translation (for example, Unraid is effectively RAID4 with some extra magic to better support mismatched drive sizes)

        That actually IS an important thing to understand early on. Because, depending on the RAID model you use, it might not be as easy as adding another drive. Have three 8 TB and want to add a 10? That last 2 TB won’t be used until EVERY drive has at least 10 TB. There are ways to set this up in ZFS and Ceph and the like but it can be a headache.

        And the issue isn’t the cloudflare tunnel. The issue is that you would have a publicly accessible service running on your network. If you use the cloudflare access control thing (login page before you can access the site) you mitigate a lot of that (while making it obnoxious for anything that uses an app…) but are still at the mercy of cloudflare.

        And understand that these are all very popular tools for a reason. So they are also things hackers REALLY care about getting access to. Just look up all the MANY MANY MANY ransomware attacks that QNAP had (and the hilarity of QNAP silently re-enabling online services with firmware updates…). Because using a botnet to just scan a list of domains and subdomains is pretty trivial and more than pays for itself after one person pays the ransom.

        As for paying for that? I would NEVER pay for nextcloud. It is fairly shit software that is overkill for what people use it for (file syncing and document server) and dogshit for what it pretends to be (google docs+drive). If I am going that route, I’ll just use Google Docs or might even check out the Proton Docs I pay for alongside my email and VPN.

        But for something self hosted where the only data that matters is backed up to a completely different storage setup? I still don’t like it being “exposed” but it is REALLY nice to have a working shopping list and the like when I head to the store.