That's cheap indeed. Enough headroom for some failing disks too. How is the noise and power usage? I didn't look at SAS drives at all, because my impression was that they're very noisy. I can place my NAS in a closed off room, but it's not too far away and I was afraid SAS drives would be audible through the wall. At the same time, the shucked drive I'm using presents itself as an WD Ultrastar, which comes very close to a SAS drive, and isn't very noisy.

The myth, popularized by a notorious thread on the TrueNAS forums [1], is specifically that ZFS requires ECC RAM, and will do worse than other filesystems without it, because scrubbing will multiply a single bitflip into a failed pool.

A ZFS core developer says that that isn't the case [2]. Here's some more reasoning [3], also about many other myths.

[1] https://www.truenas.com/community/threads/ecc-vs-non-ecc-ram...

[2] https://news.ycombinator.com/item?id=18480016

[3] https://kldload.com/zfs-wiki/myths

I'm not entirely sure, but it seems to me that free space (and the 20% reservation) is mostly a proxy for fragmentation, and you can therefore better look at fragmentation directly. That would mean that if you mostly store large files, there shouldn't be a lot of fragmentation even at high utilization. The whole "ZFS changes allocation algorithm from 80% usage on" is something of 10+ years distant past, and lots of things around the allocator have been improved. It's also something that probably isn't too different from the performance of other filesystems at high utilization, so it shouldn't be exaggerated.

I ended up on shucking 4x the 14 TB WD Elements Desktop. They contain helium drives, the WD140EDGZ in my case, and are about a third cheaper than 4x the 12 TB WD Red Plus drives (which are air-filled). The shucking was easier than I expected too, and the performance seems very comparable. The warranty is a definite downside (European, so no Magnuson-Moss), but I think I can even get them back in their enclosure should they fail during the 2-year warranty period.

I've put some second hand 256 GB M.2 SSDs in there as boot drives. It was a bit of a struggle to get it to work in a way that failure of one of the drives doesn't hold up booting, combined with LUKS, TPM keys and ZFS on root. Learned a lot about systemd-boot which I have never used before, but feels a lot saner to me than grub ever was. So now I have a large script which debootstraps a Debian based NAS into being.

I noticed that there are a lot of ZFS myths and cargo culting. For example TFA mentions ECC RAM, which in some circles is a must-have because ZFS would wreck your pool during a scrub otherwise, which is a myth. It's also very expensive, especially this year. You also don't need much RAM for ZFS, L2ARC doesn't use much RAM at all, to name a few others.

Still doubting about setting `dnodesize=auto` (which is the default), because there are some horror stories about that [1]. And it seems impossible to find a cloud storage provider with reasonable prices that supports `zfs send`. Rsync.net upped their minimum order to 10 TiB recently, which is far too much for my use case.

[1] https://github.com/openzfs/zfs/issues/11353

[2] https://www.rsync.net/products/zfsintro.html

Comments URL: https://news.ycombinator.com/item?id=48745823

Points: 1

# Comments: 0

Technical questions are unfortunately hit or miss. I'm lately pretty much always using a system prompt that emphasizes short answers [1], and Opus regularly one-shots it while Mistral needs a follow up. I use big-AGI as a model router [2] (dumb name, great software), which makes switching midway very easy though. For coding I'm still using Claude Code mostly out of inertia (although I really want to move to an OSS harness) and the one time I tried their `vibe` tool months ago it was a bit rough.

Mistral TTS with diarization is also great and cheap. That's the only thing for which I use their web UI.

[1] Give a short but helpful answer to the question the user asks. When helping with a computer-related task, unless the user asks, don't give any installation or setup instructions, but just get straight to the point. When the user asks a follow up question, give a more complete and longer answer while still not overexplaining. When the user prefaces the question with "short mode off" in any question, give a full and well considered reply.

[2] https://github.com/enricoros/big-AGI

There doesn't seem to be much written about enforcing the Data Act, so I looked at the regulation directly. Article 39 [2] seems to require to first lodge a complaint with the competent authority as designated by the member state of your residence. Then when that authority invariably fails to act – I have no idea which timeframe we're talking about here – you can "in accordance with national law, either have the right to an effective judicial remedy or access to review by an impartial body with the appropriate expertise". But then you are suing that authority, and not the company directly (edit: I was originally unsure about who to sue under article 39, but 39(3) does clarify that it is the authority).

I would very much like to be wrong about this. I can imagine Muñoz vs. Superior Fruiticola applies [3] ("it must be possible to enforce that obligation by means of civil proceedings"), but I'm not at all sure, and it's a much weaker route than the one which the GDPR explicitly describes.

Would anyone know or have better references on how to enforce the Data Act, preferably individually?

[1] https://gdpr-info.eu/art-79-gdpr/

[2] https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:...

[3] https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELE...

The activity monitor does show all kinds of Electron apps active, on top of a presumably model-loaded Handy and a virtual machine for Claude Code, so I guess that's the real root cause for all the swapping. If your laptop starts trashing I can't imagine you have any use for those apps, which will grind to a halt.

[1] https://huggingface.co/mlx-community/gemma-4-31b-it-4bit

I frequently debug issues while keeping my carefully curated but long context active for days. Losing potentially very important context while in the middle of a debugging session resulting in less optimal answers, is costing me a lot more money than the cache misses would.

In my eyes, Claude Code is mainly a context management tool. I build a foundation of apparent understanding of the problem domain, and then try to work towards a solution in a dialogue. Now you tell me Anthrophic has been silently breaking down that foundation without telling me, wasting potentially hours of my time.

It's a clear reminder that these closed-source harnesses cannot be trusted (now or in the future), and I should find proper alternatives for Claude Code as soon as possible.

[1] https://code.claude.com/docs/en/changelog

IIRC the last public exploit for all LG TVs for webOS > 5 was in the beginning of 2025 (so pretty recent), but as most sellers on the second hand market have auto-updates turned on, there's no way to know which TVs are vulnerable.

It should be doable to strip down much of webOS with root access. It's nice that webOS in general is very well documented and much is implemented around the Luna service bus. LG offers a developer mode for non-rooted TVs, and there's an active homebrew community because of it. It's a pity that you can't modify the boot partitions, as the firmware verifies their integrity. It would be nice to have an exploit for that.

[1] https://github.com/throwaway96/dejavuln-autoroot

[2] https://cani.rootmy.tv

> pip3 install git+https://github.com/ml-explore/mlx-lm.git

> ./venv/bin/mlx_lm.generate --model "$MODEL" --temp 1.0 --top-p 0.95 --top-k 64 --max-tokens 128000 --prompt "Hello world"

Where $MODEL is an unsloth model like:

- unsloth/gemma-4-E4B-it-UD-MLX-4bit

- unsloth/gemma-4-26b-a4b-it-UD-MLX-4bit

As you have so much RAM I would suggest running Q8_0 directly. It's not slower (perhaps except for the initial model load), and might even be faster, while being almost identical in quality to the original model.

And just to be sure: you're are running the MLX version, right? The mlx-community quantization seemed to be broken when I tried it last week (it spit out garbage), so I downloaded the unsloth version instead. That too was broken in mlx-lm (it crashed), but has since been fixed on the main branch of https://github.com/ml-explore/mlx-lm.

I unfortunately only have 16 GiB of RAM on a Macbook M1, but I just tried to run the Q8_0 GGUF version on a 2023 AMD Framework 13 with 64 GiB RAM just using the CPU, and that works surprisingly well with tokens/s much faster than I can read the output. The prompt cache is also very useful to quickly insert a large system prompt or file to datamine although there are probably better ways to do that instead of manually through a script.

Some models of inverter brands like Victron (which isn't very common outside its niche of self-sufficiency because they are rather expensive and sometimes complex) can form a micro-grid. They have the option of a special circuit breaker [1] that decouples the inverter from the grid if the grid is detected to be down, which allows their use during a power outage.

[1] https://www.victronenergy.com/accessories/anti-islanding-box...

[1] https://nextcloud.com/partners/

And don't get me wrong: there's nothing wrong with starting a business rebranding Nextcloud and keeping your development closed source, as long as you're honest about that, which this initiative is not.

If you're looking for a Nextcloud hoster, there's a long list of partners here [2] that have contractually obligated themselves to contribute back to Nextcloud for every user they onboard.

[1] https://blog.tomaszdunia.pl/officeeu-eng/

[2] https://nextcloud.com/partners/

Comments URL: https://news.ycombinator.com/item?id=47223610

Points: 2

# Comments: 0