Optimise ZFS performance for databases?

[helloha]

Hi,

I am using Freenas for my backups which are mostly large video files. The performance is very good.

However I also have to backup original RAW video files which are lots of individual frames of 2 MB each.

Next to that, I also have 500 GB of lightroom picture databases.

Moving large files around is fast.

However the small files or databases move very slow. Has this to do with the sector sizes?

Is it possible to optimize this?

Thanks!
K.

 

[Robert Trevellyan]

What is your pool/vdev layout?

 

[anodos]

Hi,

I am using Freenas for my backups which are mostly large video files. The performance is very good.

However I also have to backup original RAW video files which are lots of individual frames of 2 MB each.

Next to that, I also have 500 GB of lightroom picture databases.

Moving large files around is fast.

However the small files or databases move very slow. Has this to do with the sector sizes?

Is it possible to optimize this?

Thanks!
K.

Post full hardware specs. What file sharing protocol are you using? If CIFS, post /etc/local/smb4.conf
Moving around small files will always be slower than large files. Increasing the amount of VDEVs in your pool will typically improve performance.

 

[helloha]

Hi, right now i'm doing internal testing over ssh. Making copies from one pool to another in different configs.

I'm now have 2 pools. each is 4 disks in stripe. Just to test what the max is I could get and to spot the bottleneck.

I assume there is no way to set the block size in ZFS?

K.

 

[helloha]

Did some more testing. Reading from my Z1 pool and writing to my Stripe-test pool.

Consecutive, large files of 2GB each gives me 450Mb/s.

Directories of small files of 2 MB each gives me 60-80Mb/s.

All drives are 7200 rpm.

Code:

NAME                                            STATE     READ WRITE CKSUM

    HITACHI-6x3TB-14TB                              ONLINE       0     0     0

      raidz1-0                                      ONLINE       0     0     0

        gptid/42c9f476-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

        gptid/452ed1d4-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

        gptid/4786fd19-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

        gptid/49c16bb0-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

        gptid/4c0d0b53-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

        gptid/4e6bafc3-51a4-11e5-9633-001f2954783c  ONLINE       0     0     0

Code:

    NAME                                          STATE     READ WRITE CKSUM

    STRIPE-TEST-2                                 ONLINE       0     0     0

      gptid/fcf2b004-5dfd-11e5-b554-001f2954783c  ONLINE       0     0     0

      gptid/fd7ddbfa-5dfd-11e5-b554-001f2954783c  ONLINE       0     0     0

      gptid/fdf09de2-5dfd-11e5-b554-001f2954783c  ONLINE       0     0     0

      gptid/fe610214-5dfd-11e5-b554-001f2954783c  ONLINE       0     0     0

 

[cyberjock]

scp will always be slower than using things like NFS, CIFS, and FTP.

Small files will also always be slower than single large files.

 

[helloha]

Sorry, caused some confusion. I'm running all pools on one server. Just using the ssh shell to cp things from one pool to another. Not over the network. Everything internal.

 

[anodos]

Is your system a supermicro X7? The FSB might bottleneck performance.

 

[SweetAndLow]

hardware spec would be nice? The default zfs settings should be all the tuning you need to do unless you have much higher spec machine than what is average.