All of lore.kernel.org
 help / color / mirror / Atom feed
From: Qu Wenruo <quwenruo.btrfs@gmx.com>
To: Johannes Thumshirn <Johannes.Thumshirn@wdc.com>,
	Qu Wenruo <wqu@suse.com>,
	"linux-btrfs@vger.kernel.org" <linux-btrfs@vger.kernel.org>
Subject: Re: [RFC ONLY 3/8] btrfs: read raid-stripe-tree from disk
Date: Thu, 19 May 2022 21:26:54 +0800	[thread overview]
Message-ID: <e0ba76bd-72c0-fa6e-212f-92e060d79d42@gmx.com> (raw)
In-Reply-To: <PH0PR04MB741660F84BFA0F9C4E0204A79BD09@PH0PR04MB7416.namprd04.prod.outlook.com>



On 2022/5/19 19:53, Johannes Thumshirn wrote:
> On 19/05/2022 13:48, Qu Wenruo wrote:
>>
>> On 2022/5/19 19:44, Johannes Thumshirn wrote:
>>> On 19/05/2022 12:37, Qu Wenruo wrote:
>>>>> RAID1 on zoned only needs a stripe tree for data, not for meta-data/system,
>>>>> so it will work and we can bootstrap from it.
>>>>>
>>>> That sounds good.
>>>>
>>>> And in that case, we don't need to put stripe tree into system chunks at
>>>> all.
>>>>
>>>> So this method means, stripe tree is only useful for data.
>>>> Although it's less elegant, it's much saner.
>>> Yes and no. People still might want to use different metadata profiles than
>>> RAID1.
>> For RAID1 variants like RAID1C3/4, I guess we don't need stripe tree either?
>>
>> What about DUP? If RAID1*/DUP/SINGLE all doesn't need stripe tree, I
>> believe that's already a pretty good profile set for most zoned device
>> users.
>>
>> Personally speaking, it would be much simpler to avoid bothering the
>> stripe tree for metadata.
>
> I totally agree, but once you get past say 10 drives you might want to have
> different encoding schemes and also have a higher level of redundancy for your
> metadata than just 4 copies.
>
> The stripe tree will also hold any l2p information for erasure coded RAID
> arrays once that's done.
>
> So this definitively should be considered.


Then let us consider the extra chunk type flag, like
BTRFS_BLOCK_GROUP_HAS_STRIPE_TREE, and then expand the combination from
the initial RAID1*|HAS_STRIPE_TREE to other profiles.

But for over 10 devices, I doubt we really need to bother metadata that
much. Consider we go RAID1C4, we have the ability to lose 3 devices
already, that's way stronger than RAID6. For metadata I believe it's
completely fine already.

Normally it's data requiring more balance between cost and redundancy as
they are the main part of a fs.

Thus even for 10 disk, metadata RAID1C4, data RAID6 (with stripe tree
for zoned), it still looks very reasonable to me at least.

Thanks,
Qu

  reply	other threads:[~2022-05-19 13:27 UTC|newest]

Thread overview: 88+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-05-16 14:31 [RFC ONLY 0/8] btrfs: introduce raid-stripe-tree Johannes Thumshirn
2022-05-16 14:31 ` [RFC ONLY 1/8] btrfs: add raid stripe tree definitions Johannes Thumshirn
2022-05-17  7:39   ` Qu Wenruo
2022-05-17  7:45     ` Johannes Thumshirn
2022-05-17  7:56       ` Qu Wenruo
2022-05-16 14:31 ` [RFC ONLY 2/8] btrfs: move btrfs_io_context to volumes.h Johannes Thumshirn
2022-05-17  7:42   ` Qu Wenruo
2022-05-17  7:51     ` Johannes Thumshirn
2022-05-17  7:58       ` Qu Wenruo
2022-05-17  8:01         ` Johannes Thumshirn
2022-05-16 14:31 ` [RFC ONLY 3/8] btrfs: read raid-stripe-tree from disk Johannes Thumshirn
2022-05-17  8:09   ` Qu Wenruo
2022-05-17  8:13     ` Johannes Thumshirn
2022-05-17  8:28       ` Qu Wenruo
2022-05-18 11:29         ` Johannes Thumshirn
2022-05-19  8:36           ` Qu Wenruo
2022-05-19  8:39             ` Johannes Thumshirn
2022-05-19 10:37               ` Qu Wenruo
2022-05-19 11:44                 ` Johannes Thumshirn
2022-05-19 11:48                   ` Qu Wenruo
2022-05-19 11:53                     ` Johannes Thumshirn
2022-05-19 13:26                       ` Qu Wenruo [this message]
2022-05-19 13:49                         ` Johannes Thumshirn
2022-05-19 22:56                           ` Qu Wenruo
2022-05-20  8:27                             ` Johannes Thumshirn
2022-05-16 14:31 ` [RFC ONLY 4/8] btrfs: add boilerplate code to insert raid extent Johannes Thumshirn
2022-05-17  7:53   ` Qu Wenruo
2022-05-17  8:00   ` Qu Wenruo
2022-05-17  8:05     ` Johannes Thumshirn
2022-05-17  8:09       ` Qu Wenruo
2022-05-16 14:31 ` [RFC ONLY 5/8] btrfs: add code to delete " Johannes Thumshirn
2022-05-17  8:06   ` Qu Wenruo
2022-05-17  8:10     ` Johannes Thumshirn
2022-05-17  8:14       ` Qu Wenruo
2022-05-17  8:20         ` Johannes Thumshirn
2022-05-17  8:31           ` Qu Wenruo
2022-05-16 14:31 ` [RFC ONLY 6/8] btrfs: add code to read " Johannes Thumshirn
2022-05-16 14:55   ` Josef Bacik
2022-05-16 14:31 ` [RFC ONLY 7/8] btrfs: zoned: allow zoned RAID1 Johannes Thumshirn
2022-05-16 14:31 ` [RFC ONLY 8/8] btrfs: add raid stripe tree pretty printer Johannes Thumshirn
2022-05-16 14:58 ` [RFC ONLY 0/8] btrfs: introduce raid-stripe-tree Josef Bacik
2022-05-16 15:04   ` Johannes Thumshirn
2022-05-16 15:10     ` Josef Bacik
2022-05-16 15:47       ` Johannes Thumshirn
2022-05-17  7:23 ` Nikolay Borisov
2022-05-17  7:31   ` Qu Wenruo
2022-05-17  7:41     ` Johannes Thumshirn
2022-05-17  7:32   ` Johannes Thumshirn
2022-07-13 10:54 ` RAID56 discussion related to RST. (Was "Re: [RFC ONLY 0/8] btrfs: introduce raid-stripe-tree") Qu Wenruo
2022-07-13 11:43   ` Johannes Thumshirn
2022-07-13 12:01     ` Qu Wenruo
2022-07-13 12:42       ` Johannes Thumshirn
2022-07-13 13:47         ` Qu Wenruo
2022-07-13 14:01           ` Johannes Thumshirn
2022-07-13 15:24             ` Lukas Straub
2022-07-13 15:28               ` Johannes Thumshirn
2022-07-14  1:08             ` Qu Wenruo
2022-07-14  7:08               ` Johannes Thumshirn
2022-07-14  7:32                 ` Qu Wenruo
2022-07-14  7:46                   ` Johannes Thumshirn
2022-07-14  7:53                     ` Qu Wenruo
2022-07-15 17:54                     ` Goffredo Baroncelli
2022-07-15 19:08                       ` Thiago Ramon
2022-07-16  0:34                         ` Qu Wenruo
2022-07-16 11:11                           ` Qu Wenruo
2022-07-16 13:52                             ` Thiago Ramon
2022-07-16 14:26                               ` Goffredo Baroncelli
2022-07-17 17:58                                 ` Goffredo Baroncelli
2022-07-17  0:30                               ` Qu Wenruo
2022-07-17 15:18                                 ` Thiago Ramon
2022-07-17 22:01                                   ` Qu Wenruo
2022-07-17 23:00                           ` Zygo Blaxell
2022-07-18  1:04                             ` Qu Wenruo
2022-07-15 20:14                       ` Chris Murphy
2022-07-18  7:33                         ` Johannes Thumshirn
2022-07-18  8:03                           ` Qu Wenruo
2022-07-18 21:49                         ` Forza
2022-07-19  1:19                           ` Qu Wenruo
2022-07-21 14:51                             ` Forza
2022-07-24 11:27                               ` Qu Wenruo
2022-07-25  0:00                             ` Zygo Blaxell
2022-07-25  0:25                               ` Qu Wenruo
2022-07-25  5:41                                 ` Zygo Blaxell
2022-07-25  7:49                                   ` Qu Wenruo
2022-07-25 19:58                               ` Goffredo Baroncelli
2022-07-25 21:29                                 ` Qu Wenruo
2022-07-18  7:30                       ` Johannes Thumshirn
2022-07-19 18:58                         ` Goffredo Baroncelli

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=e0ba76bd-72c0-fa6e-212f-92e060d79d42@gmx.com \
    --to=quwenruo.btrfs@gmx.com \
    --cc=Johannes.Thumshirn@wdc.com \
    --cc=linux-btrfs@vger.kernel.org \
    --cc=wqu@suse.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.