Hi, I wanted to start selfhosting and I’d like to have your opinion on something that I’m struggling to decide.

I don’t plan to tinker too much with my system, I’ve been a Linux tinkerer myself some time ago but now I’d like to setup something that’s really bulletproof and then leave it running (ofc I know I’ll have to do a bit of bugfixing now and then), not replacing hardware ideally for >= 10 years.

This is why I’m planning to use TrueNAS, and that’s why I’m planning to buy a UGREEN DXP2800: has two 3,5" HDD bays (4TB should be enough for me for the next 8-10 years, so I’ll have two 4TB disks in RAID1 or mirror or whatever is recommended). Only problem I have with this machine is that it only has 1 RAM slot, and I guess 8 GB isn’t enough if I use zfs. So I’ll have to upgrade to either 16 or 32 GB. Now I did my research and from my understanding 16GB seems to be enough, but it would be such a waste having to replace the whole RAM if it turns out it isn’t enough.

For reference, I don’t plan on having more than 7-8 services running: Immich, Nextcloud+office, firefly, audiobookshelf, paperless and a maybe few more if they’re useful. I value responsiveness but it’s ok if some things take longer to process (thinking immich ML, or stuff like transcoding)

I’m very interested to know your opinion:

  • is the dxp2800 a good choice?
  • should I go with 16 or 32 GB RAM?

And a little extra

  • how much ssd space do you recommend for high speed data? is 500gb enough?

Thank you so much!

  • JASN_DE@feddit.orgEnglish
    1·
    4 hours ago

    Your only choice would be a mirror, as you only have two bays. RAIDZ2 needs at least 4 drives to be useful.

    Have you checked that you can actually run a different OS on the UGREEN NAS?

    Considering the services you want to run on the machine you’ll want all the RAM you can get.

    Also, the CPU is already weak by today’s standard, especially with “heavier” workloads like Immich machine learning.

      • bordam@feddit.itOPEnglish
        1·
        4 hours ago

        That’s great! What software do you use it for? How does it perform?

        • Tolookah@discuss.tchncs.deEnglish
          1·
          4 hours ago

          I’ve got the arr stack, nextcloud and immich, with 16gb ram and I’m pretty sure I shouldn’t add anything else.

          • bordam@feddit.itOPEnglish
            1·
            4 hours ago

            That’s actually fire, do you mean that you shouldn’t add anything more services (ram is full) or that you shouldn’t add any more ram? Anyways sounds good, I won’t be using the arr stack probably so adding a few more services won’t hurt I guess…. Is performance great? Is it responsive?

              • bordam@feddit.itOPEnglish
                1·
                39 minutes ago

                Alright, as others have said, ZFS tends to use as much ram as it’s available, so it might also be because of that. If you want to expand on the

                I wouldn’t say performance is great

                I’m all ears!!

    • bordam@feddit.itOPEnglish
      1·
      4 hours ago

      Sorry, I meant RAIDZ1 (which is the same as mirror I think?) I’ll edit the post. Anyways, Yes, I did my research and I can run TrueNAS; I plan to buy a 128gb m2 ssd for the boot device.

      Ok, so I need plenty of RAM. Do you have an idea of what the impact of something like 16 vs 32 gb of ram would be? Sluggish performance on apps would be a problem for me, but I can accept if tasks take longer.

      Got it about the CPU. I have been wondering about that too. But I heard people running immich on worse CPUs and they say the only problem is that initial ML tasks take longer (couple of days tipically), but once they’re done you’re good.

      Do you have another build to suggest in the same/slightly higher price range that you think would better suit my needs?

      tysm

      • bluGill@fedia.io
        1·
        42 minutes ago

        RaidZ1 is not the same as a mirror. I’m not sure if you are allowed to have Z1 with only 2 disks, but if you are you still shouldn’t because while it scales down that far it still does parity calculations and writes that to the second disk instead of just writing a copy of the data (the parity calculations probably result in the same data, but I doubt this is optimized)

        • bordam@feddit.itOPEnglish
          1·
          13 minutes ago

          Yeah, you’re right, I thought RAIDZ1 was ZFS’ RAID1 implementation (🤦‍♂️), while it’s actually called mirror. I’ll just do a mirror.

      • JASN_DE@feddit.orgEnglish
        2·
        4 hours ago

        Nothing besides a mirror makes sense with only 2 storage bays. The overhead would only increase for absolutely no gain.