Bcachefs Still Being Developed As A Next-Gen Linux File-System

[pal666]

RAID1 --> 3 drives wasted, 6TB array
RAID5 --> 1 drive wasted, 10TB array

To have a RAID1 with 10TB of capacity I'd need to use 10 of those drives, which means I need also another controller card, and a bigger case, and a bigger PSU and more space so on.

with raid1 you don't need controller, so it will be cheaper, faster and more robust. raid5 it made sense in the eighties

[starshipeleven]

it is enough to compare implemented features

btrfs is cooking since last decade, bcache was done in like a year or so and does not even target the same thing. No it does not make sense to compare their features.

i.e. exactly zero.

linux is mostly used in servers, so I suspect that bcache is seeing more use than whatever fs a distro pushes.

btw, btrfs is a few features on top of block layer

Every filesystem is.

and btrfs magically becomes worse than non-existent raid5 bcachefs

Never said that.

[starshipeleven]

with raid1 you don't need controller

Nor with raid5/6. mdadm can do raid5/6 too. I need sata ports tho, so I need a sata controller card. Not raid controllers, sata controllers.

Still not explaining why I should buy 4 more drives, a bigger case, a sata controller card, and a bigger PSU instead of going with raid5 in my example.

Ah, as you go higher with array size it gets worse.

EDIT: on windows you still need RAID controllers even for RAID1 (technically it has software raid but it's not that reliable)

[ssokolow]

Still not explaining why I should buy 4 more drives, a bigger case, a sata controller card, and a bigger PSU instead of going with raid5 in my example.

https://web.archive.org/web/20160112115539/http://www.miracleas.com/BAARF/RAID5_versus_RAID10.txt

[starshipeleven]

https://web.archive.org/web/20160112...sus_RAID10.txt

Uhm, we were comparing RAID1 to RAID5. And anyway RAID5 is dead with current HDD capacity https://www.thedatacave.com/raid-5-is-dead

RAID6 adds another drive and allows you to survive 2 failures, and this puts it in the "master race" vs RAID10 where you can survive 2 failures (of the right disks) or 1 failure (of the wrong disks).

But RAID6 has total shit performance, also rebuild times are a joke, while RAID10 is basically a safer RAID1, you still waste half your disks for mostly dual parity, which is dumb.

The only way to go beyond this inflexible RAID block-level shit from the eighties is having filesystems manage their own RAID.

For example, btrfs (but probably also ZFS) knows where is data and where is metadata in its array so read/writes during rebuild are just what it is needed (i.e. the blocks actually used by data or metadata are copied over), and there are no reads to disk sectors that aren't involved in this at all (so no 1 read per disk like raid5, nor syncing around free space too, which is dumb)

https://btrfs.wiki.kernel.org/index...._btrfs_raid.3F

And again the same principle applies to btrfs RAID1, it's not a RAID1 in the common sense of the word where EACH AND EVERY DISK HAS THE SAME EXACT CLONE OF THE SAME EXACT DATA, stuff is spread around the available space in two copies.
It's more similar to RAID5 in practice and space gains, allows you to place uneven amounts of disks, and use disks with different capacities (within reason, it still must be able to clone stuff on 2 places, so you can make a RAID1 with a 2TB drive and 2x 1TB drives, and it will be fine) even if it does not have striping. That's what it can pull off just by managing its own multi-drive arrays.

[nomadewolf]

still whining you two?
You can get answers from him thanks to the advanced tecnology of 21st century called "e-mail". Here you can find contact info, man up and send him a mail: https://bcache.evilpiepirate.org/#index4h1

Thanks for the insight, but...
This is a news website, so i come here that i don't have to dig around the web to find answers myself. Makes sense?
Now, at my job, when i claim that some aproach to solve a problem, i explain why. Otherwise, my boss would just tell me to fuk off.
That's how things work.
I don't make the rules.
But i do follow them.

[starshipeleven]

Thanks for the insight, but...
This is a news website, so i come here that i don't have to dig around the web to find answers myself. Makes sense?

Yes. What does not make sense is whining about lack of info, repeatedly.
If you need that info you go and ask him, or you stfu as he does not see what you post here.

[nomadewolf]

Yes. What does not make sense is whining about lack of info, repeatedly.
If you need that info you go and ask him, or you stfu as he does not see what you post here.

I just checked out his Patreon link and it does answer most of my complaints.
Also, it does look promising.
Let's wait and see.

[gilboa]

raid in fs is better, but i don't understand why someone wants 5/6. it is raid1 for poor people, slower and less robust. drive space is cheap.

1. I'd very much like to see you build a 200TB* array w/ RAID1 (as opposed to grouped RAID 50/60) **. Should be amusing.
2. RAID1 is *definitely* *less* *robust* than RAID6. Let try doing some math. Say I've got 12 x 4TB drives and I need at least 24TB of usable space. With RAID1, I need to create an array of 6 two-drive sets and somehow link them together. In theory, I can lose 6 drives out of array *as-long* as I don't lose two drives out of the same two-drive set. With RAID6 I can create a RAID with 8 active drives + 4 hot spares, lose *any* 2 drives at any given moment and 6 throughout the life of the array. Lets continue, shall we? Why not create RAID60 w/ two sets of 5+1 RAID6 array. Not only you can lose 4 drivers simultaneously (two out of each set), it'll also run circles around RAID1 in *both* read speed *and* write speed.

.... But sure, RAID5/6 is RAID1 for poor people. Right.

- Gilboa
* E.g. 2U HP Apollo 4200 w/ 28 x 8TB drives.
** Granted, such setups are usually done w/ hardware, RAIDs, but as others pointed out, a FS based RAID w/ checksum has is more capable than a dumb hardware. Plus, I do have a couple of 30-40+TB software RAIDs.

[gilboa]

But RAID6 has total shit performance, also rebuild times are a joke.

Not really. High-end (hardware) RAID controllers have no issues getting decent performance out of large multiple group RAID60 arrays. (E.g. HP P4/840, LSI 22xx, etc).

- Gilboa