logoalt Hacker News

bathtub365yesterday at 10:36 PM1 replyview on HN

The price for that storage system will be far more dominated by drive prices than by the cost of the NAS box itself. Drive prices have approximately doubled in my area vs. 2 years ago.

This is also generally a selfish attitude where you personally benefit while structures that used to benefit society at large are eroded.


Replies

FDETalkDotComtoday at 6:18 AM

> price for that storage system will be far more dominated by drive prices...doubled in my area vs. 2 years ago.

Absolutely. To get 1/8 PB = 125 TB home library "easily":

We'll use 8 disks in the 125TB library. Between RAID 5 (1 disk lost OK to recover) vs RAID 6 (2 disks lost OK to recover), choose RAID 6 (our disks could fail at same time if of similar production or unlucky). RAID6 means 25% of space used for parity overhead, and 2-5% used for metadata/filesystem.

So looking for about 163TB. 163TB / 8 rounds to 21 TB. This pushes us above 16TB disks. Between 20TB and 22TB, choose 22Tb to feel safe.

Napkin math:

  Synology 8-bay DS: $1150 (Amazon price)

  8x 22TB Seagate 22TB external 3.5" = 8 x $390 = $3120 (also the #1 least expensive disk per TB for 3.5" external at https://diskprices.co currently)
So we're at $1150+$3120 = $4270 for one library.

But something cvan happen to that. Fall, fire, water, theft, party. We could lose everything.

So following 3-2-1 we'll have 3 copies, on 2 media, with one offsite.

Copy 2 can be same as first (RAID is for disk redundancy not backup -- we still have one copy only).

By now, 2x Synology 8-bays, plus 16x 22TB disks, puts us at $8540 for what we can keep at home.

But disks only really last about 5 years. They're getting kinda better, but in reality those disks can fail and should be replaced about every 5 years, some people get 10.

So every 5 years, we can want to shell ou;t about $8540. But wait, disks about doubled in the past year. Maybe it'll be $16,000 next time? Hard to say.

We still need a 3rd, off-site copy for 3-2-1. Recent reports indicated Backblaze silently lost data, some people exodused I believe. To where? IDK, but let's pick Amazon Glacier deep storage. At 125TB (just useful data), at $0.00099/GB/mo, that puts it at, over the same 5 years: $0.00099/GB/mo * 125000 GB * 12mo * 5yr = $7,425/5yr

(For the remote copy: can your ISP actually handle uploading 125TB? How long does that take to do once, even half? Is ISP transfer capped? Will 3rd party storage provider change prices or lose data? That's why we have 3 copies, maybe change providers when needed.

In any case, add it on 3-2-1 for 125TB would cost, at the easiest/cheapest: $4270 * 2 + $7,425 = 4270+7425 = $15,965, good for about 5 years.

So, every 5 years, spending $15,965.

At these volumes, are do even have ECC RAM? Are we scanning for and correcting errors with correct data when they occur? We don't want a hobby, we want an appliance, for this library, often especially if we work 99% of the time in tech and have life to live, quite likely.

Let's try another formula: on a shoestring and a hope, one could do it "cheap on RAID 5 (only 12.5% lost to parity and metadata/filesystem) and under-storage without 3-2-1" by going Synology 8-bay ($1150, Amazon) + 8 * 16TB (8 * $410 per https://diskprices.co = $3280) = $1150 + $3280 = $4430

---

  In grand summary, roughly every 5 years:

  Done "right": $15,965
  Done "cheap": $4,430 and only 112TB usable.
You know what, 112TB starts to feel like not that much, when we look at the size of some of the libraries out there.

Averaged over 5 years (though it's not) these are:

  - Right: $15,965 / 5yr = about $3,200/yr (plus tax) for 125TB usable library
  - Cheap: $4,430 / 5yr = $886/yr (plus tax) for 112TB usable library
If a techie makes $150K, that's about 0.6%-2% of income, if we forget taxes (sales or income) entirely.

Maybe doable. But it's like owning another car in more ways than one (cost, maintenance/ongoing-care). Some individuals can swing it without even thinking. Most can't.

IMO, if the AI industry or any players would like us to become more computer centric, and make use of all the data that tech now lets us have, its constituents should do something (anything) to drive the cost of disks DOWN, not UP.