NetApp and Solaris

Solaris NetApp Just some unsorted lines... Working on it... don't believe what you can read here! It is not proofed for now.

Settings in Solaris

Settings for MPxIO over FC:

/kernel/drv/ssd.conf

###### START changes by host_config #####
ssd-config-list="NETAPP  LUN", "physical-block-size:4096, retries-busy:30, retries-reset:30, retries-notready:300, retriestimeout:10, throttle-max:64, throttle-min:8";
###### END changes by host_config ####

Check it out

# iostat -Er | /opt/sfw/bin/gawk 'BEGIN{command="echo ::ssd_state | mdb -k"; while(command|getline){if(/^un [0-9]+:/ && $NF != "0"){ssd=$2;gsub(/:$/,"",ssd);while(!/^}/){command|getline;if(/un_phy_blocksize/){un_phy_blocksize[ssd]=strtonum($NF);}}}};close(command);}/ssd/{ssd=$1;gsub(/^ssd/,"",ssd);getline;split($0,vendor,",");printf "ssd: %s\tun_phy_blocksize: %d\t%s\t%s\n",ssd,un_phy_blocksize[ssd],vendor[1],vendor[4];}'

Alignment and ZFS

First read ZFS zpools create misaligned I/O in Solaris 11 and Solaris 10 Update 8 and later (407376).

If you have 4k as block size in your storage use ashift=12 (alignment shift exponent).

Status of alignment

# ssh filer01 "priv set -q diag ; lun show -v all; priv set"
-------------------------------------------------------------------------------
LUN for ZFS
-------------------------------------------------------------------------------
        /vol/ZoneLUNs/Zone01.lun      50g (53687091200)   (r/w, online, mapped)
                Serial#: 800KP+EpO-33
                Share: none
                Space Reservation: enabled
                Multiprotocol Type: solaris_efi
                Maps: SUN_SERVER01_SERVER02=40
                Occupied Size:   46.2g (49583595520)  
                Creation Time: Wed Jan  7 11:37:58 CET 2015
--->            Alignment: partial-writes
                Cluster Shared Volume Information: 0x0 
                Space_alloc: disabled
                report-physical-size: enabled
                Read-Only: disabled
-------------------------------------------------------------------------------
LUN for Oracle Database
-------------------------------------------------------------------------------
        /vol/TEMP201/TEMP201          25g (26843545600)   (r/w, online, mapped)
                Serial#: 800KP+EpO-2t
                Share: none
                Space Reservation: enabled
                Multiprotocol Type: solaris_efi
                Maps: KSP_SUN_SRV06_SRV07=35
                Occupied Size:   21.6g (23195856896)  
                Creation Time: Fri Jul  4 11:02:34 CEST 2014
--->            Alignment: misaligned
                Cluster Shared Volume Information: 0x0 
                Space_alloc: disabled
                report-physical-size: enabled
                Read-Only: disabled
...

Or use "lun alignment show":

# ssh filer01 "priv set -q diag ; lun alignment show; priv set"
-------------------------------------------------------------------------------
LUN for ZFS
Wide spread reads. I think the ashift is not correct.
-------------------------------------------------------------------------------
        /vol/ZoneLUNs/Zone01.lun 
                Multiprotocol type: solaris_efi
                Alignment: partial-writes
                Write alignment histogram percentage: 5, 5, 4, 6, 4, 6, 14, 5
                Read alignment histogram percentage: 8, 7, 10, 7, 7, 8, 36, 5
                Partial writes percentage: 47
                Partial reads percentage: 9
-------------------------------------------------------------------------------
LUN for Oracle Database
-------------------------------------------------------------------------------
        /vol/TEMP201/TEMP201     
                Multiprotocol type: solaris_efi
                Alignment: misaligned
                Write alignment histogram percentage: 0, 0, 0, 0, 0, 0, 99, 0
                Read alignment histogram percentage: 0, 0, 8, 0, 0, 0, 77, 0
                Partial writes percentage: 0
                Partial reads percentage: 14

Or "stats show lun":

filer01*> stats show -e lun:/vol/TEMP201:.*_align_histo.*

ashift=12? Why 12?

# echo "2^12" | bc -l
4096

OK... 4k... I see.

What ashift do I have?

# zdb | egrep ' name|ashift'
    name: 'apache_pool'
            ashift: 9
    name: 'mysql_pool'
            ashift: 9
...

Create ZPools on NetApp LUNs with this syntax

# zpool create -o ashift=12 <mypool> mirror <vdev1> <vdev2>

Links

• Illumos: List of sd-config-list entries for Advanced-Format drives
• NetApp: What is an unaligned I/O?