All of lore.kernel.org
 help / color / mirror / Atom feed
* what should blockdev getbpsz return 4k disks w/512e?
@ 2017-11-04  9:09 L A Walsh
  2017-11-06 10:19 ` Karel Zak
  0 siblings, 1 reply; 4+ messages in thread
From: L A Walsh @ 2017-11-04  9:09 UTC (permalink / raw)
  To: util-linux


I just tried blockdev on a HW raid device (/dev/sdb) composed of
4K disks that can (w/performance penalty) emulate 512 byte disk
(i.e. have the 512e feature).  It is returning 512. 

If the disk is composed of 4K disks, shouldn't the raid show 4K
as well?  Otherwise, how do utils know what alignment to use or
not to use 512 due to penalty costs?  For my formatted disks, I
use 4K allocation units, so for those, I shouldn't see problems,
but some other disk subsystems or raw-disk using progs also need to
know to allocate on 4K boundaries.

Any idea where this info should be showing up?

Thanks!
-linda


^ permalink raw reply	[flat|nested] 4+ messages in thread
* Re: what should blockdev getbpsz return 4k disks w/512e?
  2017-11-04  9:09 what should blockdev getbpsz return 4k disks w/512e? L A Walsh
@ 2017-11-06 10:19 ` Karel Zak
  2017-11-14  4:37   ` L A Walsh
  0 siblings, 1 reply; 4+ messages in thread
From: Karel Zak @ 2017-11-06 10:19 UTC (permalink / raw)
  To: L A Walsh; +Cc: util-linux

On Sat, Nov 04, 2017 at 02:09:51AM -0700, L A Walsh wrote:
> 
> I just tried blockdev on a HW raid device (/dev/sdb) composed of
> 4K disks that can (w/performance penalty) emulate 512 byte disk
> (i.e. have the 512e feature).  It is returning 512.
> 
> If the disk is composed of 4K disks, shouldn't the raid show 4K
> as well?  Otherwise, how do utils know what alignment to use or
> not to use 512 due to penalty costs?  For my formatted disks, I
> use 4K allocation units, so for those, I shouldn't see problems,
> but some other disk subsystems or raw-disk using progs also need to
> know to allocate on 4K boundaries.
> 
> Any idea where this info should be showing up?

Try

 --getiomin                get minimum I/O size
 --getioopt                get optimal I/O size

or

  lsblk --topology

It's possible that RAID (or another stacked device) uses 512
addressing, but for partitioning or mkfs-like tools we care about
minimal and optimal I/O.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

^ permalink raw reply	[flat|nested] 4+ messages in thread
* Re: what should blockdev getbpsz return 4k disks w/512e?
  2017-11-06 10:19 ` Karel Zak
@ 2017-11-14  4:37   ` L A Walsh
  2017-11-14 10:08     ` Karel Zak
  0 siblings, 1 reply; 4+ messages in thread
From: L A Walsh @ 2017-11-14  4:37 UTC (permalink / raw)
  To: Karel Zak; +Cc: util-linux

[-- Attachment #1: Type: text/plain, Size: 2022 bytes --]

Karel Zak wrote:
> On Sat, Nov 04, 2017 at 02:09:51AM -0700, L A Walsh wrote:
>   
>> I just tried blockdev on a HW raid device (/dev/sdb) composed of
>> 4K disks that can (w/performance penalty) emulate 512 byte disk
>> (i.e. have the 512e feature).  It is returning 512.
>>
>> Any idea where this info should be showing up?
>>     
>
> Try
>
>  --getiomin                get minimum I/O size
>  --getioopt                get optimal I/O size
>   
----
 
  > sudo blockdev --getiomin /dev/sdb
  512
  > sudo blockdev --getioopt /dev/sdb
  0
> or
>
>   lsblk --topology
>   
-----
(some lines deleted for brevity):

NAME                 ALIGN IO:MIN OPT  PHYSEC LOGSEC ROT SCHED RQSZ  RA  
WSAME
sda                    0    512    0     512    512   1   bfq  256   
512    0B
└─sda1                 0    512    0     512    512   1   bfq  256   
512    0B
  └─Backup-Media       0    512    0     512    512   1        128   
128    0B
sdb                    0    512    0     512    512   1   bfq  256   
512    0B
└─sdb1                 0    512    0     512    512   1   bfq  256   
512    0B
  ├─Data-Home          0    512    0     512    512   1        128   
128    0B
  └─Data-Edge          0    512    0     512    512   1        128   
128    0B
sdc                    0    512    0     512    512   1   bfq  256   
512    0B
├─sdc1                 0    512    0     512    512   1   bfq  256   
512    0B
└─sdc10                0    512    0     512    512   1   bfq  256   
512    0B

> It's possible that RAID (or another stacked device) uses 512
> addressing, but for partitioning or mkfs-like tools we care about
> minimal and optimal I/O.
>   
----
    Nothing but the RAID controller seems to know about the 4096 
physical size.
    (attached jpg from disk controller SW -- can see size under physical 
on right). 

    If anything used a 512 byte offset anywhere, it would really impact 
performance, thus my concern about how or where a util would be able to read
it...






[-- Attachment #2: disk_sizes1.jpg --]
[-- Type: image/jpeg, Size: 175005 bytes --]

^ permalink raw reply	[flat|nested] 4+ messages in thread
* Re: what should blockdev getbpsz return 4k disks w/512e?
  2017-11-14  4:37   ` L A Walsh
@ 2017-11-14 10:08     ` Karel Zak
  0 siblings, 0 replies; 4+ messages in thread
From: Karel Zak @ 2017-11-14 10:08 UTC (permalink / raw)
  To: L A Walsh; +Cc: util-linux

On Mon, Nov 13, 2017 at 08:37:37PM -0800, L A Walsh wrote:
> Karel Zak wrote:
> > On Sat, Nov 04, 2017 at 02:09:51AM -0700, L A Walsh wrote:
> > > I just tried blockdev on a HW raid device (/dev/sdb) composed of
> > > 4K disks that can (w/performance penalty) emulate 512 byte disk
> > > (i.e. have the 512e feature).  It is returning 512.
> > > 
> > > Any idea where this info should be showing up?
> > 
> > Try
> > 
> >  --getiomin                get minimum I/O size
> >  --getioopt                get optimal I/O size
> ----
> 
>  > sudo blockdev --getiomin /dev/sdb
>  512
>  > sudo blockdev --getioopt /dev/sdb
>  0
> > or
> > 
> >   lsblk --topology
> -----
> (some lines deleted for brevity):
> 
> NAME                 ALIGN IO:MIN OPT  PHYSEC LOGSEC ROT SCHED RQSZ  RA
> WSAME
> sda                    0    512    0     512    512   1   bfq  256   512
> 0B
> └─sda1                 0    512    0     512    512   1   bfq  256   512
> 0B
>  └─Backup-Media       0    512    0     512    512   1        128   128
> 0B
> sdb                    0    512    0     512    512   1   bfq  256   512
> 0B
> └─sdb1                 0    512    0     512    512   1   bfq  256   512
> 0B
>  ├─Data-Home          0    512    0     512    512   1        128   128
> 0B
>  └─Data-Edge          0    512    0     512    512   1        128   128
> 0B
> sdc                    0    512    0     512    512   1   bfq  256   512
> 0B
> ├─sdc1                 0    512    0     512    512   1   bfq  256   512
> 0B
> └─sdc10                0    512    0     512    512   1   bfq  256   512
> 0B
> 
> > It's possible that RAID (or another stacked device) uses 512
> > addressing, but for partitioning or mkfs-like tools we care about
> > minimal and optimal I/O.
> ----
>    Nothing but the RAID controller seems to know about the 4096 physical
> size.
>    (attached jpg from disk controller SW -- can see size under physical on
> right).
> 
>    If anything used a 512 byte offset anywhere, it would really impact
> performance, thus my concern about how or where a util would be able to read
> it...

Unfortunately userspace tools cannot play any usable role in this
story if your system (kernel) does not provide relevant information.
For mkfs and partitioning tools it's 512-bytes based devices... 

Let's ask kernel developers.

    Karel

-- 
 Karel Zak  <kzak@redhat.com>
 http://karelzak.blogspot.com

^ permalink raw reply	[flat|nested] 4+ messages in thread
end of thread, other threads:[~2017-11-14 10:08 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-04  9:09 what should blockdev getbpsz return 4k disks w/512e? L A Walsh
2017-11-06 10:19 ` Karel Zak
2017-11-14  4:37   ` L A Walsh
2017-11-14 10:08     ` Karel Zak

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.