Differences in su/sw values for hw vs. sw RAID 5?

Differences in su/sw values for hw vs. sw RAID 5?

[bridavis]

I getting conflicting reports as to how I should generate my sunit/swidth vaules for hardware RAID 5.

Setup: hardware RAID 5, 3 disks at 300 GBs each, 64k stripe size.

Originally, following the man page and the mailing list archives, I came up sw=2,su=64k. 

However, I read a reply to an earlier question I sent to the list, and it indicated that the hardward RAID should be treated as a single disk, so I came up with sw=1,su=128k.

Which one is correct for my setup?

Thanks!

[[HTML alternate version deleted]]

Re: Differences in su/sw values for hw vs. sw RAID 5?

[Shailendra Tripathi]

For RAID-5 device, for any write, the parity as well has to be 
calculated before writing. In absence of any column of RAID, it is read 
from disk and then re-written. When you choose writes such as all 
columns are already there, parity can be directly calculated and written 
(without incurring any extra read I/O) and that's why, declaring in that 
form is desirable. Someone correct me if I am wrong.

# mdadm --create /dev/md15 --level=5 --raid-devices=3 -c 64 /dev/sd[hvi]1
mdadm: array /dev/md15 started.

When forced choice of sw=1,su=128k
# cat /proc/mdstat | more
...
md15 : active raid5 sdv1[2] sdi1[1] sdh1[0]
       78139904 blocks level 5, 64k chunk, algorithm 2 [3/3] [UUU]
# mkfs.xfs -f -d sw=1,su=128k /dev/md15
mkfs.xfs: Specified data stripe unit 256 is not the same as the volume 
stripe unit 128
meta-data=/dev/md15              isize=256    agcount=16, agsize=1220928 
blks
          =                       sectsz=512
data     =                       bsize=4096   blocks=19534848, imaxpct=25
          =                       sunit=32     swidth=32 blks, unwritten=1
naming   =version 2              bsize=4096
log      =internal log           bsize=4096   blocks=9568, version=1
          =                       sectsz=512   sunit=0 blks
realtime =none                   extsz=131072 blocks=0, rtextents=0

Though by default, it detects the former one.

# mkfs.xfs -f /dev/md15
meta-data=/dev/md15              isize=256    agcount=16, agsize=1220944 
blks
          =                       sectsz=512
data     =                       bsize=4096   blocks=19534976, imaxpct=25
          =                       sunit=16     swidth=32 blks, unwritten=1
naming   =version 2              bsize=4096

Please note that default created here is: sunit=16, swidth=3
bridavis@comcast.net wrote:
> I getting conflicting reports as to how I should generate my sunit/swidth vaules for hardware RAID 5.
> 
> Setup: hardware RAID 5, 3 disks at 300 GBs each, 64k stripe size.
> 
> Originally, following the man page and the mailing list archives, I came up sw=2,su=64k. 
> 
> However, I read a reply to an earlier question I sent to the list, and it indicated that the hardward RAID should be treated as a single disk, so I came up with sw=1,su=128k.
> 
> Which one is correct for my setup?
> 
> Thanks!
> 
> [[HTML alternate version deleted]]
> 
> 

Re: Differences in su/sw values for hw vs. sw RAID 5?

[Brian Davis]

Maybe I'm missing something, but I'm not sure how the information below 
maps to setting the values on Hardware RAID.

A nice feature of xfs is that it's intelligent enough to figure out the 
proper values for SW RAID.

Thanks!

Shailendra Tripathi wrote:
> For RAID-5 device, for any write, the parity as well has to be 
> calculated before writing. In absence of any column of RAID, it is 
> read from disk and then re-written. When you choose writes such as all 
> columns are already there, parity can be directly calculated and 
> written (without incurring any extra read I/O) and that's why, 
> declaring in that form is desirable. Someone correct me if I am wrong.
>
> # mdadm --create /dev/md15 --level=5 --raid-devices=3 -c 64 /dev/sd[hvi]1
> mdadm: array /dev/md15 started.
>
> When forced choice of sw=1,su=128k
> # cat /proc/mdstat | more
> ...
> md15 : active raid5 sdv1[2] sdi1[1] sdh1[0]
>       78139904 blocks level 5, 64k chunk, algorithm 2 [3/3] [UUU]
> # mkfs.xfs -f -d sw=1,su=128k /dev/md15
> mkfs.xfs: Specified data stripe unit 256 is not the same as the volume 
> stripe unit 128
> meta-data=/dev/md15              isize=256    agcount=16, 
> agsize=1220928 blks
>          =                       sectsz=512
> data     =                       bsize=4096   blocks=19534848, imaxpct=25
>          =                       sunit=32     swidth=32 blks, unwritten=1
> naming   =version 2              bsize=4096
> log      =internal log           bsize=4096   blocks=9568, version=1
>          =                       sectsz=512   sunit=0 blks
> realtime =none                   extsz=131072 blocks=0, rtextents=0
>
> Though by default, it detects the former one.
>
> # mkfs.xfs -f /dev/md15
> meta-data=/dev/md15              isize=256    agcount=16, 
> agsize=1220944 blks
>          =                       sectsz=512
> data     =                       bsize=4096   blocks=19534976, imaxpct=25
>          =                       sunit=16     swidth=32 blks, unwritten=1
> naming   =version 2              bsize=4096
>
> Please note that default created here is: sunit=16, swidth=3
> bridavis@comcast.net wrote:
>> I getting conflicting reports as to how I should generate my 
>> sunit/swidth vaules for hardware RAID 5.
>>
>> Setup: hardware RAID 5, 3 disks at 300 GBs each, 64k stripe size.
>>
>> Originally, following the man page and the mailing list archives, I 
>> came up sw=2,su=64k.
>> However, I read a reply to an earlier question I sent to the list, 
>> and it indicated that the hardward RAID should be treated as a single 
>> disk, so I came up with sw=1,su=128k.
>>
>> Which one is correct for my setup?
>>
>> Thanks!
>>
>> [[HTML alternate version deleted]]
>>
>>
>