Bug #954

Discussion:

Bug #954

nils privat via illumos-zfs

2014-07-21 19:54:29 UTC

Hello, i have a question about the Illumos Ticket #954

https://www.illumos.org/issues/954

Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.

I would be really great if you can solve that "inefficient space problem"
with 4K Disks in raidz-mode.

Thanks,
gismo3303

-------------------------------------------
illumos-zfs
Archives: https://www.listbox.com/member/archive/182191/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182191/23047029-187a0c8d
Modify Your Subscription: https://www.listbox.com/member/?member_id=23047029&id_secret=23047029-2e85923f
Powered by Listbox: http://www.listbox.com

Matthew Ahrens via illumos-zfs

2014-07-21 20:38:24 UTC

Permalink

I don't understand that bug report. It's talking about 4KB block sizes.
If you aren't already using the largest block sizes available (128KB), why
would you want blocks even larger than that?

That said, if you think it's a dup of 5027, feel free to close it as such.

--matt

On Mon, Jul 21, 2014 at 12:54 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.
I would be really great if you can solve that "inefficient space problem"
with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify
<https://www.listbox.com/member/?&>
Your Subscription <http://www.listbox.com>

-------------------------------------------
illumos-zfs
Archives: https://www.listbox.com/member/archive/182191/=now
RSS Feed: https://www.listbox.com/member/archive/rss/182191/23047029-187a0c8d
Modify Your Subscription: https://www.listbox.com/member/?member_id=23047029&id_secret=23047029-2e85923f
Powered by Listbox: http://www.listbox.com

Timothy Coalson via illumos-zfs

2014-07-21 21:15:15 UTC

Permalink

I think it is referring to the 4K sector disk problem (ashift=12), and its
effect on the total usable size of the pool.

An old thread that has a table of this overhead:

http://permalink.gmane.org/gmane.os.illumos.zfs/1951

Tim

On Mon, Jul 21, 2014 at 3:38 PM, Matthew Ahrens via illumos-zfs <

Post by Matthew Ahrens via illumos-zfs
I don't understand that bug report. It's talking about 4KB block sizes.
If you aren't already using the largest block sizes available (128KB), why
would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt
On Mon, Jul 21, 2014 at 12:54 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.
I would be really great if you can solve that "inefficient space problem"
with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/24077749-f7a4e87a> |
Modify
<https://www.listbox.com/member/?&>
Your Subscription <http://www.listbox.com>

Matthew Ahrens via illumos-zfs

2014-07-21 21:19:03 UTC

Permalink

I didn't get that from reading the bug report. However:

See http://blog.delphix.com/matt/2014/06/06/zfs-stripe-width/
Click through to the goolgle spreadsheet. Note that sizes are in sectors,
so with 4KB sectors, there are currently at most 32 sectors per block. If
going from 32 sectors to 256 would help utilization for your RAID-Z stripe
size, then you could benefit from 1MB block size vs 128KB.

--matt

Post by Timothy Coalson via illumos-zfs
I think it is referring to the 4K sector disk problem (ashift=12), and its
effect on the total usable size of the pool.
http://permalink.gmane.org/gmane.os.illumos.zfs/1951
Tim
On Mon, Jul 21, 2014 at 3:38 PM, Matthew Ahrens via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.
I would be really great if you can solve that "inefficient space
problem" with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

Richard Yao via illumos-zfs

2014-07-21 21:20:59 UTC

Permalink

Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to 16KB. Storing a zvol with the default 8KB volblocksize will result in the zvol using twice the actual space when it is filled.

To answer the original question, no one is working on this right now.

I don't understand that bug report. It's talking about 4KB block sizes. If you aren't already using the largest block sizes available (128KB), why would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request large blok size.
I would be really great if you can solve that "inefficient space problem" with 4K Disks in raidz-mode.
Thanks,
gismo3303
illumos-zfs | Archives | Modify Your Subscription

illumos-zfs | Archives | Modify Your Subscription

nils privat via illumos-zfs

2014-07-21 21:26:22 UTC

Permalink

Yes, i mean that big overhead problem with 4K (ashift=12) and raidz.

See:

https://github.com/zfsonlinux/zfs/issues/548
or
http://comments.gmane.org/gmane.os.illumos.devel/11004
or
http://hardforum.com/archive/index.php/t-1784350.html

2014-07-21 23:20 GMT+02:00 Richard Yao via illumos-zfs <

Post by Richard Yao via illumos-zfs
Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to
16KB. Storing a zvol with the default 8KB volblocksize will result in the
zvol using twice the actual space when it is filled.
To answer the original question, no one is working on this right now.
On Jul 21, 2014, at 4:38 PM, "Matthew Ahrens via illumos-zfs" <
I don't understand that bug report. It's talking about 4KB block sizes.
If you aren't already using the largest block sizes available (128KB), why
would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt
On Mon, Jul 21, 2014 at 12:54 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request large blok size.
I would be really great if you can solve that "inefficient space problem"
with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/24010604-91e32bd2> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/26373437-199173a4> |
Modify
<https://www.listbox.com/member/?&>
Your Subscription <http://www.listbox.com>

Matthew Ahrens via illumos-zfs

2014-07-21 21:31:27 UTC

Permalink

On Mon, Jul 21, 2014 at 2:26 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Yes, i mean that big overhead problem with 4K (ashift=12) and raidz.

I think you mean, "and small block sizes" (e.g. zvols with the default 8KB
volblocksize, or small files).

--matt

Post by nils privat via illumos-zfs
https://github.com/zfsonlinux/zfs/issues/548
or
http://comments.gmane.org/gmane.os.illumos.devel/11004
or
http://hardforum.com/archive/index.php/t-1784350.html
2014-07-21 23:20 GMT+02:00 Richard Yao via illumos-zfs <
Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to

Post by Richard Yao via illumos-zfs
16KB. Storing a zvol with the default 8KB volblocksize will result in the
zvol using twice the actual space when it is filled.
To answer the original question, no one is working on this right now.
On Jul 21, 2014, at 4:38 PM, "Matthew Ahrens via illumos-zfs" <
I don't understand that bug report. It's talking about 4KB block sizes.
If you aren't already using the largest block sizes available (128KB), why
would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt
On Mon, Jul 21, 2014 at 12:54 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.
I would be really great if you can solve that "inefficient space
problem" with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/24010604-91e32bd2> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/26373437-199173a4> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify
<https://www.listbox.com/member/?&>
Your Subscription <http://www.listbox.com>

Matthew Ahrens via illumos-zfs

2014-07-21 21:47:16 UTC

Permalink

Post by Richard Yao via illumos-zfs
Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to
16KB.

I don't understand that. Blocks on RAIDZ-2 will be padded to be a multiple
of 3 sectors (12KB), including the parity. See the table I mentioned for
details; look at the "RAIDZ-2 Parity cost" tab, the column with 6 disks.

https://docs.google.com/a/delphix.com/spreadsheets/d/1tf4qx1aMJp8Lo_R6gpT689wTjHv6CGVElrPqTA0w_ZY/edit?pli=1#gid=930519344

Post by Richard Yao via illumos-zfs
Storing a zvol with the default 8KB volblocksize will result in the zvol
using twice the actual space when it is filled.

Actually, it should use 3x the space that it would use on a non-redundant
pool, because it will need 6 sectors: 2 data + 2 parity + 2 to round up to
a multiple of 3 (P+1).

--matt

Post by Richard Yao via illumos-zfs
To answer the original question, no one is working on this right now.
On Jul 21, 2014, at 4:38 PM, "Matthew Ahrens via illumos-zfs" <
I don't understand that bug report. It's talking about 4KB block sizes.
If you aren't already using the largest block sizes available (128KB), why
would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt
On Mon, Jul 21, 2014 at 12:54 PM, nils privat via illumos-zfs <

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request large blok size.
I would be really great if you can solve that "inefficient space problem"
with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>

Richard Yao via illumos-zfs

2014-07-21 23:24:11 UTC

Permalink

Post by Richard Yao via illumos-zfs
Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to 16KB.

There was a change to the code at some point to pad it to a full stripe to avoid situations where disks are underutilized. There was a comment elaborating on this when I last looked.

https://docs.google.com/a/delphix.com/spreadsheets/d/1tf4qx1aMJp8Lo_R6gpT689wTjHv6CGVElrPqTA0w_ZY/edit?pli=1#gid=930519344

Post by Richard Yao via illumos-zfs
Storing a zvol with the default 8KB volblocksize will result in the zvol using twice the actual space when it is filled.

Actually, it should use 3x the space that it would use on a non-redundant pool, because it will need 6 sectors: 2 data + 2 parity + 2 to round up to a multiple of 3 (P+1).
--matt

`zfs list` will display space usage as doubling if recall my empirical tests from a few years ago correctly, but you are right.

Post by Richard Yao via illumos-zfs
To answer the original question, no one is working on this right now.

I don't understand that bug report. It's talking about 4KB block sizes. If you aren't already using the largest block sizes available (128KB), why would you want blocks even larger than that?
That said, if you think it's a dup of 5027, feel free to close it as such.
--matt

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request large blok size.
I would be really great if you can solve that "inefficient space problem" with 4K Disks in raidz-mode.
Thanks,
gismo3303
illumos-zfs | Archives | Modify Your Subscription

illumos-zfs | Archives | Modify Your Subscription

Matthew Ahrens via illumos-zfs

2014-07-23 04:37:59 UTC

Permalink

On Jul 21, 2014, at 5:47 PM, "Matthew Ahrens via illumos-zfs" <

Post by Richard Yao via illumos-zfs
Objects stored on a raidz2 ashift=12 vdev with 6 disks will be padded to 16KB.

I don't understand that. Blocks on RAIDZ-2 will be padded to be a
multiple of 3 sectors (12KB), including the parity. See the table I
mentioned for details; look at the "RAIDZ-2 Parity cost" tab, the column
with 6 disks.
There was a change to the code at some point to pad it to a full stripe to
avoid situations where disks are underutilized. There was a comment
elaborating on this when I last looked.

I don't recall that. If you can point me to the code/comment, I'd be
curious to see it.

https://docs.google.com/a/delphix.com/spreadsheets/d/1tf4qx1aMJp8Lo_R6gpT689wTjHv6CGVElrPqTA0w_ZY/edit?pli=1#gid=930519344

Post by Richard Yao via illumos-zfs
Storing a zvol with the default 8KB volblocksize will result in the zvol
using twice the actual space when it is filled.

Actually, it should use 3x the space that it would use on a non-redundant
pool, because it will need 6 sectors: 2 data + 2 parity + 2 to round up to
a multiple of 3 (P+1).
`zfs list` will display space usage as doubling if recall my empirical
tests from a few years ago correctly, but you are right.

That makes sense. "zfs list" (and "du" / "ls -s", but not "zpool list")
take into account the estimated RAID-Z parity overhead. The estimate is
based on 128KB block size -- 50% in the case of a 6-wide RAID-Z2. The
actual space used for 8KB blocks is twice this estimate: data + 50% parity
= *150%* vs data + 200% parity = *300%*.

--matt

Post by nils privat via illumos-zfs
Hello, i have a question about the Illumos Ticket #954
https://www.illumos.org/issues/954
Do you have any Idea when that Ticket will be in Progress?
A few Days ago there was submitted a new Ticket #5027 which also request
large blok size.
I would be really great if you can solve that "inefficient space
problem" with 4K Disks in raidz-mode.
Thanks,
gismo3303
*illumos-zfs* | Archives
<https://www.listbox.com/member/archive/182191/=now>
<https://www.listbox.com/member/archive/rss/182191/21635000-ebd1d460> |
Modify <https://www.listbox.com/member/?&> Your Subscription
<http://www.listbox.com>