Mastodawn

raspberry Dec 21

oh wow, anna's archive "backed up" Spotify! they downloded 256 million songs - roughly 300TB - so you could technically, make your own spotify at home! would easily fit in a 2RU server, hmmmmm

https://annas-archive.li/blog/backing-up-spotify.html

Backing up Spotify

We backed up Spotify (metadata and music files). It’s distributed in bulk torrents (~300TB). It’s the world’s first “preservation archive” for music which is fully open (meaning it can easily be mirrored by anyone with enough disk space), with 86 million music files, representing around 99.6% of listens.

@decryption what’s the payback period on this? Haha

Irenes (many)Dec 20

@decryption you probably need 6U to get 300T with current drive sizes, if you find a hardware setup that can do it in less physical space we'd love to hear about it

but yes, that is well within reach of anyone who wants to spend the money

decryption Dec 20

@ireneista 12x 30TB HDDs = 360TB of space - not accounting for RAID or anything, just one big sloppy JBOD

Irenes (many)Dec 20

@decryption right but where are you seeing 30T drives on the market? 20T was the high end of the price/size curve the last time we looked, and it's nonlinear enough that we're not sure we'd advise anyone to actually buy that size (as opposed to more drives in a slightly smaller drive)

Irenes (many)Dec 20

@decryption it has been a little while since we seriously priced out a large storage purchase though, so maybe we're out of date

Wade Roberts Dec 20

@ireneista @decryption https://www.scorptec.com.au/product/hard-drives-&-ssds/hdd-3.5-drives/119922-st30000nt011

Irenes (many)Dec 20

@waderoberts @decryption huh, neat. we didn't realize things had gotten that far.

our point about nonlinear pricing stands, that's a lot for a single drive, but it does seem like someone who really wanted to spend the money could do that.

decryption Dec 20

@ireneista yeah 28TB HDDs are bang for buck now

https://east-digital.myshopify.com/

East Digital

East Digital

East Digital

Irenes (many)Dec 20

@decryption the largest drive that store is offering right now is 16T? are we reading it wrong?

decryption Dec 20

@ireneista yes lol

https://east-digital.myshopify.com/products/st28000nm000c-seagate-exos-28tb-3-5-sata-7200rpm-256mb-enterprise-hard-drive-1

ST28000NM000C Seagate Exos 28TB 3.5" SATA 7200RPM 512MB Enterprise Hard Drive

Irenes (many)Dec 20

@decryption thanks. sorry for the noise.

Irenes (many)Dec 20

@decryption sorry - we were reading it wrong, we see a 28T there now. we're unsure why it didn't show in the full list.

@ireneista @decryption If one is alright with getting re-certified drives, you can snag some 28TB Exos disks for just under the 15 USD/TB mark at $410 per drive - https://serverpartdeals.com/products/seagate-exos-st28000nm000c-28tb-7-2k-rpm-sata-6gb-s-512e-cmr-3-5-recertified-hard-drive

Otherwise you're probably looking at that 16-20TB range for the lowest dollar per TB cost - a new 18TB drive on SPD runs as much as the recertified 28TB drive (sure you can get a new 20TB drive for even cheaper on SPD but it's SMR so uh, not great)

(If one adds in a RAID setup, 18 drives can hold at minimum 252TB with 28TB drives in a RAID10 setup, swapping to parity boosts that a fair bit, even if you just do RAID50 or RAID60 to parity clusters. Whether that's smart or not IDK)

Seagate Exos ST28000NM000C 28TB 7.2K RPM SATA 6Gb/s 512e CMR 3.5in Recertified Hard Drive

Irenes (many)Dec 20

@senil @decryption yeah that second range you mention is what we were thinking in terms of, because most people are going to treat price as, at least, somewhere in their priorities

neofox_peek_owo

@decryption @ireneista
https://www.fractal-design.com/products/cases/define/define-7-xl/black-solid/

Up to 18 HDDs in this case, I use it as my primary server and have ~120TB usable with ZFS over 9 x 20TB HDDs.

18 x 28TB = 504TB maxed out in JBOD
11 x 28TB = 308TB as JBOD (using BTRFS or SnapRaid I guess, without much parity too).

I guess on it's side the Define 7 XL would be a 12U ? Or a 4U, I'm not really sure.

Define 7 XL

The Define 7 XL sets a new standard for what you should expect from a full tower case in terms of modularity, flexibility and ease of use.

Fractal Design

Cassandrich Dec 20

@ireneista @decryption Micro SD cards come up to 512GB now, no? 600 of those would be tiny. Put them all on one SPI bus with chipselect lines.

Cassandrich Dec 20

@ireneista @decryption The beauty with this kind of data is you don't need high transfer rates or concurrency, so you can optimize for low cost storage at rest for low physical space & energy usage.

Haelwenn /элвэн/

🔜FOSDEM Dec 20

@dalias @ireneista @decryption Reminds me of tapes robots kind of stuff, and seems to be lowest price per gig with still a decent density. (And 0 power draw when unused)

dr2chase Dec 20

@dalias @ireneista @decryption you'd want a little bit of RAID there, some of those will undoubtedly fail over time. I've only had medium luck with microSD cards in video cameras.

Irenes (many)Dec 20

@dalias @decryption yeah using SD cards for higher density is a really interesting idea. we have had enough data loss involving SD cards that it isn't a thing we'd personally try but it's very good to keep in mind, there are doubtless situations where it would be the right tradeoff

@ireneista @decryption @dalias We live in a world where you could eat an entire Spotify if you tried hard enough. Incredible

mx alex tax1a - 2020 (6)Dec 20

@er1 @ireneista @decryption @dalias sounds like the kind of thing that would give you indigestion

Irenes (many)Dec 20

@er1 @decryption @dalias yeah!!!!! it's really cool to realize. our digression about trying to ballpark how you'd actually do it and what's cost-effective doesn't change OP's core point. it's quite something.

Elias Mårtenson Dec 21

@er1 @ireneista @decryption @dalias I calculated the information density of punch cards once (not saying it was hard, you just need three numbers, and the only difficult part was finding a reliable measue of the thickness of a card).

I'll look it up later, but I will leave my initial guess for the size of the archive as being a volume that could fit in a fully loaded container ship.

Cassandrich Dec 20

@ireneista @decryption If they're operated read-only once they're initially populated, they might be really reliable. All the failures I've had on SD cards are from bad devices (Raspberry Pi's) with misdesigned electronics frying them on power failure during writes.

Irenes (many)Dec 20

@dalias @decryption yeah we'd want to test it before relying on it but that seems plausible

@dalias @ireneista @decryption

Divide them across a couple of stacks and you could totally make a surprisingly portable media player that way, heh.

Irenes (many)Dec 21

@remmy @dalias @decryption ahahahahaa oh that is such a delightful thought. "portable" :DDDD

Irenes (many)Dec 21

@remmy @dalias @decryption it reminds us of how, before there were laptop computers, there were luggable computers

"portable" depends heavily on who you are...

Cassandrich Dec 21

@ireneista @remmy @decryption There were towers we could carry to LAN parties. 😂

Irenes (many)Dec 21

@dalias @remmy @decryption yessssssss :D

@ireneista @dalias @decryption

Right!

But on the other hand, I think you could probably manage to make something not significantly larger than a 3.5" HDD, and even that's not really significantly bigger than a CD walkman, just arranged differently.

Cassandrich Dec 21

@remmy @ireneista @decryption MicroSD cards are 11x15x1 mm, so 165 mm³ each. 600 of them would be 99 cm³, so like a 9x11x1 cm box. I'm thinking of having them soldered to layers of polyimide (kapton) flex pcb.

Irenes (many)Dec 21

@dalias @remmy @decryption soldering them in place would definitely be smaller than those little metal insert cages, yeah

Cassandrich Dec 21

@ireneista @remmy @decryption Cheaper and easier to assemble too.

@dalias @ireneista @decryption

Seams feasible, yeah. It would mean that replacing a single broken SD card would be rather difficult, but 600 sockets would be Bonkers Expensive™.

I'd be a bit worried about the long-term life of the solder joints for a flex PCB, but maybe it would be okay if designed carefully.

@dalias @ireneista @decryption

Bonus points if the case and input methods mimic iPod Classics and you call the device something like iSpod(ify).

Cassandrich Dec 21

@remmy @ireneista @decryption You wouldn't be flexing them; the whole setup would be enclosed. The polyimide "flex pcb" is just to get thin conductive traces insulated on the other side.

If you wanted to do replacements you'd separate the layers and take off the bad card or use reflow heat to remove them all and re-solder the whole layer.

@dalias @ireneista @decryption

My worry would be that it would still flex over time from vibrations while in the enclosure.

kouhai, of the health issues Dec 21

@dalias @remmy @ireneista @decryption pondering this interesting idea: n+2 or n+3 redundancy, with fusing off individual sd “cells” as necessary

kouhai, of the health issues Dec 21

@dalias @remmy @ireneista @decryption i realize that my brain didn’t dump everything down

- instead of a single atomic system, split storage into multiple modules, each which can be individually taken offline for replacement

and/or

- build in some spare space and some parity and a block mapping layer, so you can reallocate bad sd cards entirely

Not to derail your point but I have a 1.5tb microsd in my steamdeck and I think someone announced a 2tb though I don’t know if it’s out yet.

But that cuts the number down to 150-200 sd cards.

@ireneista @decryption

disorderlyf Dec 21

@dalias @ireneista @decryption I've seen 1TB at Micro Center before. I heard someone say legit 2TB cards exist now.

EDIT: To clarify, my point is if we're talking about the viability of using SD cards for mass storage, all else being equal, you can do more with less.

gkrnours Dec 21

@disorderlyf @dalias @ireneista @decryption
https://shop.sandisk.com/en-ie/products/memory-cards/microsd-cards/sandisk-extreme-uhs-i-microsd?sku=SDSQXAV-2T00-GN6MA
https://www.ldlc.com/fiche/PB00716807.html

for 300 bucks, you can have a 2TB card

2TB SanDisk Extreme® microSDXC™ UHS-I CARD | Sandisk

Extreme speed for data transfer and app performance are guaranteed with the SanDisk Extreme microSDXC UHS-I memory card.

Sandisk

Irenes (many)Dec 21

@gkrnours @disorderlyf @dalias @decryption everything's tradeoffs. notice that this is roughly 10x more expensive per terabyte than a spinning-metal drive.

disorderlyf Dec 21

@ireneista @gkrnours @dalias @decryption In this case, you pay 10x more for 10x smaller physical size.

Irenes (many)Dec 21

@disorderlyf @gkrnours @dalias @decryption yeah, absolutely. in fact the physical size improvement is significantly better than that.

kouhai, of the health issues Dec 21

@dalias @ireneista @decryption express cards go to 1TB, and they’re “just” nvme

Lexar example:

https://www.bhphotovideo.com/c/product/1892652-REG/lexar_lmsxps0001t_bnnnu_1tb_play_pro_microsdxc.html?

You can get a petabyte raw capacity in a 4U box. Fill one of these with 18TB spinning disk and you're good to go. that'd be the cheap option

for big dollars, get one of these and fill it with 30TB ssds (this is only 720 gigs but it'd store the music collection with full redundancy)

SSG-542B-E1CR60 | 4U | SuperServer | Products | Supermicro

4U top-loading storage server with 60 hot-swap 3.5" SAS/SATA bays and PCIe 5.0

Irenes (many)Dec 20

@khm oh very nice. yeah our personal rack is not quite that deep but it's still nice to see those options.

the vessel of morganna Dec 21

@ireneista @decryption You can get 12 bays in a 2U chassis front-only, or 24 bays with top-loaders (the second set of 12 folds out from the middle of the chassis). Assuming 36T HAMRs that's 432T / 864T raw, totally within reach. Prices are mighty high for those top loader chassis though, and the HAMRs are no joke either. 18-20T drives are a quarter of the price and more classic 4U storage chassis tend to be a lot cheaper too (and mechanically simpler)

Irenes (many)Dec 21

@astraleureka @decryption yeah with hindsight our estimate of the space requirement was based on a depth constraint that we personally have but that not everybody does, and which it isn't really fair to consider in the context of this thread

Irenes (many)Dec 21

@astraleureka @decryption also we didn't say it very well but we were trying to talk less about what's theoretically possible and more about what's a good balance of cost vs. other constraints

Irenes (many)Dec 21

@astraleureka @decryption in both those respects it would have been better had we ignored the original post and made our own thread. so, sorry about that.

the vessel of morganna Dec 21

@ireneista @decryption Yeah it's totally ridiculous to go with the 36T drives right now unless your blood is 80% cash. The 16-20T range is far more affordable still, and older 12 bay chassis are dirt cheap in comparison (maybe $200-300 vs $7000 for the top loaders). I've got a 12 bay 2U for personal storage loaded with 18Ts and even that was pricy

@ireneista @astraleureka @decryption So go 4U DS4246 and fill it with something like 28TB drives? Might be cheaper to go a bit smaller drives but that'll do north of 150TB/RU raw.

benschwarz Dec 20

@decryption I get that people don’t want to pay Spotify for various reasons, but resorting to thieving music artists work ain’t it

decryption Dec 20

@benschwarz anna's approach is preservation, not so much piracy

Jonathan Hendry Dec 20

@decryption @benschwarz

Oh so it’s going to be stored offline in a salt mine?

Sure, Jan.

Cassandrich Dec 20

@benschwarz @decryption Spotify already did that. Artists don't get paid when you listen on Spotify.

To clarify: there's a minimum number of streams to get paid at all, so small artists literally get zero. Nothing. Big ones get pennies. It's not a viable source of income.