On Wed, Apr 8, 2015 at 7:02 PM, Jason Matthews <[email protected]> wrote:

>
>
> Do you have any notion what if any performance penalty there is staying
> with 512 byte sectors?
>

That's entirely hardware dependent.  Varies between "none" and "a lot" :-)

--matt


>
> Given the cost of high quality SSDs I am not inclined to move to larger
> sector sizes unless there is a real pay off. Right now, I dont see it.
>
> j.
>
>
> On 4/8/2015 5:41 PM, Matthew Ahrens wrote:
>
>  On Wed, Apr 8, 2015 at 5:12 PM, jason matthews <[email protected]> wrote:
>
>>
>> cross posting to zfs
>>
>> The data seems to indicate that something strange is happening between
>> the compression algorithms and  sector sizes > 512 bytes. ZFS reports
>> virtually the same compression ratios for either ashift=9 or ashift=12
>> zpools,
>
>
>  This is a bug.  The compression ratio with sector size other than 512
> bytes (i.e. ashift != 9) is reported higher than it really is.  I have a
> fix in the works.
>
>
>> but the number of data blocks
>
>
>  I think you mean amount of space allocated, e.g. as reported by the
> "referenced" or "used" properties.  Number of data blocks should be the
> same.
>
>
>> consumed varies drastically between 9 bit and 12 bit sectors.
>
>
>  I think you mean 512-byte and 4KB sectors (2^9 and 2^12).  9-bit sectors
> sounds like something from my nightmares :-)
>
>
>> Uncompressed file systems report the same disk usage so the problem seems
>> tied towards the compression mechanisms. Moving to 13 bit, as the illumos
>> advanced sector wiki page implies, results in even more amplication.
>>
>> why should 4k sectors consume nearly 2x storage compressed versus 512
>> byte zpools? this looks like some sort of bug to me.
>
>
> This is not a bug, it's an inherent effect of sector size.  Take for
> example a 8KB (logical size) block that compresses down to 4.12KB.  With
> 512 byte sectors, we will allocate 4.5KB (9 sectors), vs with 4KB sectors,
> we must allocate 8KB (2 sectors).  This effect will be more pronounced with
> smaller blocks (e.g. recordsize=8k), and not very noticeable with large
> blocks (e.g. the default of 128KB), unless the compression ratio is very
> high (e.g. 128KB blocks that compresses to 4.12KB)
>
>  --matt
>
>
>>
>> With Identical datasets installed on each file system we see a almost 2x
>> magnification in disk usage in ashift=12 versus ashift=9.
>
>
>> Do you any of you big brains on the ZFS team know why this happens?
>> Is this a bug? Some sort of errant behavior in the Intel DC S3700?
>> If it is not a bug, why would we want to move to 4k sectors?
>> Is this a case where ZFS compression isnt tuned against 4k blocks?
>>
>> should I file a ticket?
>>
>> root@dbb005:/root# zfs list -r ashift9 ashift12 |grep postgres
>> ashift12/128k/lz4/postgres              355G  40.7G   355G
>> /ashift12/128k/lz4/postgres
>> ashift12/128k/lzjb/postgres             403G  40.7G   403G
>> /ashift12/128k/lzjb/postgres
>> ashift12/128k/no-compression/postgres   640G  40.7G   640G
>> /ashift12/128k/no-compression/postgres
>> ashift12/8k/lz4/postgres                355G  40.7G   355G
>> /ashift12/8k/lz4/postgres
>> ashift12/8k/lzjb/postgres               403G  40.7G   403G
>> /ashift12/8k/lzjb/postgres
>>
>> ashift9/128k/lz4/postgres               224G  6.32T   224G
>> /ashift9/128k/lz4/postgres
>> ashift9/128k/lzjb/postgres              277G  6.32T   277G
>> /ashift9/128k/lzjb/postgres
>> ashift9/128k/no-compression/postgres    638G  6.32T   638G
>> /ashift9/128k/no-compression/postgres
>> ashift9/8k/lz4/postgres                 224G  6.32T   224G
>> /ashift9/8k/lz4/postgres
>> ashift9/8k/lzjb/postgres                224G  6.32T   224G
>> /ashift9/8k/lzjb/postgres
>>
>> root@dbb005:/root# zfs get -r compressratio ashift9 ashift12 | grep
>> postgres |grep -v @
>> ashift12/128k/lz4/postgres                       compressratio 2.86x  -
>> ashift12/128k/lzjb/postgres                      compressratio 2.31x  -
>> ashift12/128k/no-compression/postgres            compressratio 1.00x  -
>> ashift12/8k/lz4/postgres                         compressratio 2.86x  -
>> ashift12/8k/lzjb/postgres                        compressratio 2.31x  -
>>
>> ashift9/128k/lz4/postgres                        compressratio 2.86x  -
>> ashift9/128k/lzjb/postgres                       compressratio 2.31x  -
>> ashift9/128k/no-compression/postgres             compressratio 1.00x  -
>> ashift9/8k/lz4/postgres                          compressratio 2.86x  -
>> ashift9/8k/lzjb/postgres                         compressratio 2.86x  -
>>
>>
>>
>> root@dbb005:/root# zpool status ashift12 ashift9
>>   pool: ashift12
>>  state: ONLINE
>>   scan: none requested
>> config:
>>
>>         NAME                     STATE     READ WRITE CKSUM
>>         ashift12                 ONLINE       0     0     0
>>           c3t55CD2E404B63FCD9d0  ONLINE       0     0     0
>>           c3t55CD2E404B64033Bd0  ONLINE       0     0     0
>>           c3t55CD2E404B63F93Cd0  ONLINE       0     0     0
>>
>> errors: No known data errors
>>
>> root@dbb005:/root# zdb ashift12 |egrep ashift:
>>                 ashift: 12
>>                 ashift: 12
>>                 ashift: 12
>>                 ashift: 12
>>                 ashift: 12
>>                 ashift: 12
>>
>>
>>
>>   pool: ashift9
>>  state: ONLINE
>>   scan: none requested
>> config:
>>
>>         NAME                     STATE     READ WRITE CKSUM
>>         ashift9                  ONLINE       0     0     0
>>           c3t55CD2E404B417CFEd0  ONLINE       0     0     0
>>           c3t55CD2E404B43F0C9d0  ONLINE       0     0     0
>>           c3t55CD2E404B640592d0  ONLINE       0     0     0
>>           c3t55CD2E404B43F183d0  ONLINE       0     0     0
>>           c3t55CD2E404B417D1Fd0  ONLINE       0     0     0
>>           c3t55CD2E404B43F14Cd0  ONLINE       0     0     0
>>           c3t55CD2E404B43EFAAd0  ONLINE       0     0     0
>>           c3t55CD2E404B43EFCCd0  ONLINE       0     0     0
>>           c3t55CD2E404B41904Ad0  ONLINE       0     0     0
>>           c3t50015178F3659395d0  ONLINE       0     0     0
>>           c3t55CD2E404B64062Fd0  ONLINE       0     0     0
>>
>> errors: No known data errors
>>
>> root@dbb005:/root# zdb ashift9 |egrep ashift:
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>                 ashift: 9
>>
>> root@dbb005:/root# iostat -En
>> c3t55CD2E404B417CFEd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV331202X3800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B41904Ad0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV331207RE800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B8149FAd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2CT18 Revision: 335u Serial No:
>> CVKI338200ZT180
>> Size: 180.05GB <180045766656 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B43EFAAd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV333005CU800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B406BBCd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV330209W2800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B417D1Fd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV331202Y3800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B43F0C9d0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV333005ME800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B43EFCCd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV333005DU800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t50015178F3659395d0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV326303UP800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B43F14Cd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV333005RC800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B43F183d0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: DL04 Serial No:
>> BTTV333005T3800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B63FCD9d0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: 0270 Serial No:
>> BTTV420600LD800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B640592d0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: 0270 Serial No:
>> BTTV420602JK800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B63F93Cd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: 0270 Serial No:
>> BTTV420505XQ800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B64062Fd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: 0270 Serial No:
>> BTTV420602P6800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>> c3t55CD2E404B64033Bd0 Soft Errors: 0 Hard Errors: 0 Transport Errors: 0
>> Vendor: ATA      Product: INTEL SSDSC2BA80 Revision: 0270 Serial No:
>> BTTV4206020X800
>> Size: 800.17GB <800166076416 bytes>
>> Media Error: 0 Device Not Ready: 0 No Device: 0 Recoverable: 0
>> Illegal Request: 0 Predictive Failure Analysis: 0
>>
>> All Intel DC S3700 drives used in the ashift9 and ashift12 zpools
>>
>>
>
>
_______________________________________________
developer mailing list
[email protected]
http://lists.open-zfs.org/mailman/listinfo/developer

Reply via email to