====== NAS - Build a Linux NAS - Tune the System ====== There are many settings that can be tuned. As there are many factors than can affect tuning performance, it is recommended to perform a number of tests using different figures to determine the optimum values: * Create a test file that will be used for reading and writing tests. * Perform each test using **dd**. dd if=testfile of=/dev/null bs=8k * Between each test, clear the OS disk cache. sync;echo 3 > /proc/sys/vm/drop_caches **NOTE:** Vary the size of the test file depending on how much RAM the system has. * It is suggested to have the test file be large, such as 32G. * Consider changing the blocksize used with dd to a larger value, say bs=16k, if there is a lot of RAM. **WARNING:** Some changes may result in a bit of extra memory being used. * After each command, the amount of memory being used can be checked with: free -m returns: total used free shared buff/cache available Mem: 64326 22064 5354 1366 36907 40203 Swap: 975 315 660 ---- ===== Tuning stripe_cache_size ===== **stripe_cache_size** affects RAM used by mdadm for writing of data. Ubuntu default value is 256. Verify the current value: cat /sys/block/md0/md/stripe_cache_size Change it: echo *number* > /sys/block/md0/md/stripe_cache_size or echo *number* | sudo tee /sys/block/md0/md/stripe_cache_size **NOTE:** Test with different sizes. * Test with stripe_cache_size=256 * Test with stripe_cache_size=512 * Test with stripe_cache_size=1024 * Test with stripe_cache_size=2048 * Test with stripe_cache_size=4096 * Test with stripe_cache_size=8192 * Test with stripe_cache_size=16384 * Test with stripe_cache_size=32768 At each test, check the **Write** performance, for example: dd if=testfile of=/dev/null bs=8k 361680+0 records in 361680+0 records out 2962882560 bytes (3.0 GB) copied, 0.92471 s, 3.2 GB/s Read speed test: dd if=testfile of=/dev/zero Write speed test: dd if=/dev/zero of=testfile Also check block devices: blockdev --report ---- ===== Tuning Read Ahead ===== Changing the read ahead value should impact **read** performance. * Default read ahead value is **1536**. Change it:: blockdev --setra *number* /dev/md0 **NOTE:** Test with different sizes. * Test with Read Ahead @ 1536 * Test with Read Ahead @ 4096 * Test with Read Ahead @ 32768 * Test with Read Ahead @ 262144 * Test with Read Ahead @ 524288 At each test, check the **Read** performance, for example: dd if=testfile of=/dev/null bs=8k 361680+0 records in 361680+0 records out 2962882560 bytes (3.0 GB) copied, 0.92471 s, 3.2 GB/s Read speed test: dd if=testfile of=/dev/zero Write speed test: dd if=/dev/zero of=testfile Also check block devices: blockdev --report ---- ===== Script to auto-determine best tuning changes ===== #!/bin/bash # # Please note this test requires 30 GB of free space # in your RAID md device # # The aim of this script is to find the best settings for performance # of your RAID by testing each setting separately. # This script does make some system modification, but if you don't # make these changes permanent (e.g. write them in /etc/rc.local) # At the next boot all the changes will be lost, # so fill free to play with it! # # developed by alfonso / Jan 2012 # # # this is your mount point for the RAID # PLEASE NOTE the script will REMOVE any file called testfile*.out in this folder!! MNT=/mnt/storage # this is device from which to get input # no need to change this INPUT=/dev/zero # test for privileges if [ "$(whoami)" != 'root' ] then echo Need to be root! echo ABORT exit 1 fi if ! [ -d $MNT/lost+found ] then echo echo "$MNT is not a file system! Something went wrong?" echo ABORT exit 1 fi # find out which one is your md # note that the script only works for one md. If you have more than one # just uncomment the line below and type something like MDDEV=md0 MDDEV="`cat /proc/mdstat | grep md | head -1 | awk '{print $1}'`" # MDDEV=md0 if [ -z "$MDDEV" ] then echo echo "I can\'t find any md" echo ABORT exit 1 fi # # get the letter of all devices from cat /proc/mdstat # # this expression takes the output of /proc/mdstat # then takes the line of our md # then changes spaces into new lines # then takes only lines starting with sd # then take the 3rd character (a for sda1, etc) # and then remove new lines to make a single string DEVS="`cat /proc/mdstat | grep $MDDEV | tr " " "\n" | grep '^sd' | awk '{print substr($0,3,1)}' | tr -d "\n"`" echo "These are devices found in $MDDEV: $DEVS" function test_write() { # writing tests: echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null WRTE1=`dd if=$INPUT of=$MNT/testfile1.out bs=100kB count=100000 2>&1 | grep copied` WRUN1=`echo $WRTE1 | awk ' { print ( $(NF) ) }'` WRSP1=`echo $WRTE1 | awk ' { print ( $(NF-1) ) }'` if [ $WRUN1 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null WRTE2=`dd if=$INPUT of=$MNT/testfile2.out bs=1MB count=10000 2>&1 | grep copied` WRUN2=`echo $WRTE2 | awk ' { print ( $(NF) ) }'` WRSP2=`echo $WRTE2 | awk ' { print ( $(NF-1) ) }'` if [ $WRUN2 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null WRTE3=`dd if=$INPUT of=$MNT/testfile3.out bs=10MB count=1000 2>&1 | grep copied` WRUN3=`echo $WRTE3 | awk ' { print ( $(NF) ) }'` WRSP3=`echo $WRTE3 | awk ' { print ( $(NF-1) ) }'` if [ $WRUN3 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi AVG_WRITE=`echo "($WRSP1+$WRSP2+$WRSP3)*100/3" | bc` #echo " Average write is $AVG_WRITE MB/s NOTE: there should be a dot before the last 2 digits" echo " average write is `echo "scale=2; $AVG_WRITE / 100;" | bc` MB/s" # echo $WRTE1 # echo $WRTE2 # echo $WRTE3 } function test_read() { # reading tests: echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null READ1=`dd if=$MNT/testfile1.out of=/dev/null bs=100kB count=100000 2>&1 | grep copied` RDUN1=`echo $READ1 | awk ' { print ( $(NF) ) }'` RDSP1=`echo $READ1 | awk ' { print ( $(NF-1) ) }'` if [ $RDUN1 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null READ2=`dd if=$MNT/testfile2.out of=/dev/null bs=1MB count=10000 2>&1 | grep copied` RDUN2=`echo $READ2 | awk ' { print ( $(NF) ) }'` RDSP2=`echo $READ2 | awk ' { print ( $(NF-1) ) }'` if [ $RDUN2 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi echo -n . hdparm -f /dev/sd[$DEVS] > /dev/null READ3=`dd if=$MNT/testfile3.out of=/dev/null bs=10MB count=1000 2>&1 | grep copied` RDUN3=`echo $READ3 | awk ' { print ( $(NF) ) }'` RDSP3=`echo $READ3 | awk ' { print ( $(NF-1) ) }'` if [ $RDUN3 != "MB/s" ]; then echo echo "This script was created for all speeds measured in MB/s" echo ABORT exit 1 fi AVG_READ=`echo "($RDSP1+$RDSP2+$RDSP3)*100/3" | bc` #echo " Average read is $AVG_READ MB/s NOTE: there should be a dot before the last 2 digits" echo " average read is `echo "scale=2; $AVG_READ / 100;" | bc` MB/s" #echo $READ1 #echo $READ2 #echo $READ3 } echo echo CURRENT SYSTEM SETTINGS echo your current value of /sys/block/$MDDEV/md/stripe_cache_size is `cat /sys/block/$MDDEV/md/stripe_cache_size` echo your current value of disk readahead is `blockdev --getra /dev/sd[$DEVS]` echo your current value of md readahead is `blockdev --getra /dev/$MDDEV` DEVINDEX=0 NUMDEVS=${#DEVS} until [ $DEVINDEX -ge $NUMDEVS ] do DEVLETTER=${DEVS:$DEVINDEX:1} echo your current value of /sys/block/sd$DEVLETTER/queue/max_sectors_kb is `cat /sys/block/sd$DEVLETTER/queue/max_sectors_kb` DEVINDEX=$[$DEVINDEX+1] done echo for i in 1 2 3 4 #for i in 1 # 1 when testing /sys/block/$MDDEV/md/stripe_cache_size # 2 when testing disk readahead # 3 when testing md readahead # 4 when testing /sys/block/sdX/queue/max_sectors_kb do BEST_WRITE=0 BEST_WRITE_ID=0 WORST_WRITE=0 WORST_WRITE_ID=0 BEST_READ=0 BEST_READ_ID=0 WORST_READ=0 WORST_READ_ID=0 for j in 64 128 256 512 1024 2048 4096 8192 16384 # for j in 64 16384 do #echo #echo SYSTEM SETTINGS #echo your current value of /sys/block/$MDDEV/md/stripe_cache_size is `cat /sys/block/$MDDEV/md/stripe_cache_size` #echo your current value of disk readahead is `blockdev --getra /dev/sd[$DEVS]` #echo your current value of md readahead is `blockdev --getra /dev/$MDDEV` #DEVINDEX=0 #NUMDEVS=${#DEVS} #until [ $DEVINDEX -ge $NUMDEVS ] #do # DEVLETTER=${DEVS:$DEVINDEX:1} # echo your current value of /sys/block/sd$DEVLETTER/queue/max_sectors_kb is `cat /sys/block/sd$DEVLETTER/queue/max_sectors_kb` # DEVINDEX=$[$DEVINDEX+1] #done #echo case "$i" in 1) echo "We are testing md stripe_cache_size" echo $j > /sys/block/$MDDEV/md/stripe_cache_size echo "step 1/4: NOW your current value of /sys/block/$MDDEV/md/stripe_cache_size is `cat /sys/block/$MDDEV/md/stripe_cache_size`" ;; 2) echo "We are testing disks readahead" blockdev --setra $j /dev/sd[$DEVS] echo "step 2/4: NOW your current value of disk readahead is `blockdev --getra /dev/sd[$DEVS]`" ;; 3) echo "We are testing md readahead" blockdev --setra $j /dev/$MDDEV echo "step 3/4 NOW your current value of md readahead is `blockdev --getra /dev/$MDDEV`" ;; 4) echo "We are testing disks max_sectors_kb" DEVINDEX=0 NUMDEVS=${#DEVS} until [ $DEVINDEX -ge $NUMDEVS ] do DEVLETTER=${DEVS:$DEVINDEX:1} echo $j > /sys/block/sd$DEVLETTER/queue/max_sectors_kb echo "step 4/4 NOW your current value of /sys/block/sd$DEVLETTER/queue/max_sectors_kb is `cat /sys/block/sd$DEVLETTER/queue/max_sectors_kb`" DEVINDEX=$[$DEVINDEX+1] done ;; *) echo "This text should never appear" echo ABORT exit 1 ;; esac rm $MNT/testfile*.out 2> /dev/null test_write if [ "$BEST_WRITE" -eq "0" ] then #echo 1st test BEST_WRITE BEST_WRITE=$AVG_WRITE BEST_WRITE_ID=$j fi if [ "$WORST_WRITE" -eq "0" ] then #echo 1st test WORST_WRITE WORST_WRITE=$AVG_WRITE WORST_WRITE_ID=$j fi if [ "$AVG_WRITE" -ge "$BEST_WRITE" ] then echo "found new best write - old: `echo "scale=2; $BEST_WRITE / 100;" | bc` new: `echo "scale=2; $AVG_WRITE / 100;" | bc`" #echo "old: $BEST_WRITE new: $AVG_WRITE" BEST_WRITE=$AVG_WRITE BEST_WRITE_ID=$j fi if [ "$AVG_WRITE" -le "$WORST_WRITE" ] then echo "found new worst write - old: `echo "scale=2; $WORST_WRITE / 100;" | bc` new: `echo "scale=2; $AVG_WRITE / 100;" | bc`" #echo old: $WORST_WRITE new: $AVG_WRITE WORST_WRITE=$AVG_WRITE WORST_WRITE_ID=$j fi test_read if [ "$BEST_READ" -eq "0" ] then #echo 1st test BEST_READ BEST_READ=$AVG_READ BEST_READ_ID=$j fi if [ "$WORST_READ" -eq "0" ] then #echo 1st test WORST_READ WORST_READ=$AVG_READ WORST_READ_ID=$j fi if [ "$AVG_READ" -ge "$BEST_READ" ] then echo "found new best read - old: `echo "scale=2; $BEST_READ / 100;" | bc` new: `echo "scale=2; $AVG_READ / 100;" | bc`" #echo old: $BEST_READ new: $AVG_READ BEST_READ=$AVG_READ BEST_READ_ID=$j fi if [ "$AVG_READ" -le "$WORST_READ" ] then echo "found new worst read - old: `echo "scale=2; $WORST_READ / 100;" | bc` new: `echo "scale=2; $AVG_READ / 100;" | bc`" #echo old: $WORST_READ new: $AVG_READ WORST_READ=$AVG_READ WORST_READ_ID=$j fi rm $MNT/testfile1.out rm $MNT/testfile2.out rm $MNT/testfile3.out done echo BEST_WRITE is $BEST_WRITE echo BEST_WRITE_ID is $BEST_WRITE_ID echo WORST_WRITE is $WORST_WRITE echo WORST_WRITE_ID is $WORST_WRITE_ID echo BEST_READ is $BEST_READ echo BEST_READ_ID is $BEST_READ_ID echo WORST_READ is $WORST_READ echo WORST_READ_ID is $WORST_READ_ID # now we want to understand if this test affected more READ or WRITE performances DIFF_WRITE=$[ BEST_WRITE - WORST_WRITE ] DIFF_READ=$[ BEST_READ - WORST_READ ] if [ "$DIFF_READ" -gt "$DIFF_WRITE" ] then echo this test affected more READ than WRITE BEST_OVERALL_ID=$BEST_READ_ID WORST_OVERALL_ID=$WORST_READ_ID else echo this test affected more WRITE than READ BEST_OVERALL_ID=$BEST_WRITE_ID WORST_OVERALL_ID=$WORST_WRITE_ID fi case "$i" in 1) echo "$BEST_OVERALL_ID is the OPTIMAL value for md stripe_cache_size" BEST_1_ID=$BEST_OVERALL_ID echo $BEST_OVERALL_ID > /sys/block/$MDDEV/md/stripe_cache_size ;; 2) echo "$BEST_OVERALL_ID is the OPTIMAL value for disks readahead" BEST_2_ID=$BEST_OVERALL_ID blockdev --setra $BEST_OVERALL_ID /dev/sd[$DEVS] ;; 3) echo "$BEST_OVERALL_ID is the OPTIMAL value for md readahead" BEST_3_ID=$BEST_OVERALL_ID blockdev --setra $BEST_OVERALL_ID /dev/$MDDEV ;; 4) echo "$BEST_OVERALL_ID is the OPTIMAL value for max_sectors_kb" BEST_4_ID=$BEST_OVERALL_ID DEVINDEX=0 NUMDEVS=${#DEVS} until [ $DEVINDEX -ge $NUMDEVS ] do DEVLETTER=${DEVS:$DEVINDEX:1} echo $BEST_OVERALL_ID > /sys/block/sd$DEVLETTER/queue/max_sectors_kb DEVINDEX=$[$DEVINDEX+1] done ;; *) echo "This text should never appear" echo ABORT exit 1 ;; esac done echo the best for md stripe_cache_size is $BEST_1_ID echo the best for disks readahead is $BEST_2_ID echo the best for md readahead is $BEST_3_ID echo the best for max_sectors_kb is $BEST_4_ID echo echo "Add the following lines to your /etc/rc.local" echo echo "echo $BEST_1_ID > /sys/block/$MDDEV/md/stripe_cache_size" echo "blockdev --setra $BEST_2_ID /dev/sd[$DEVS]" echo "blockdev --setra $BEST_3_ID /dev/$MDDEV" DEVINDEX=0 NUMDEVS=${#DEVS} until [ $DEVINDEX -ge $NUMDEVS ] do DEVLETTER=${DEVS:$DEVINDEX:1} echo "echo $BEST_4_ID > /sys/block/sd$DEVLETTER/queue/max_sectors_kb" DEVINDEX=$[$DEVINDEX+1] done exit 0 ---- ===== Make these tuning changes permanent ===== There are many ways to allow this. The suggested approach is to create a file **/etc/profile.d/raid_tuning.sh**: echo *number* | sudo tee /sys/block/md0/md/stripe_cache_size blockdev --setra *number* /dev/md0 For example: echo 8192 > /sys/block/md0/md/stripe_cache_size echo 256 > /sys/block/sdb/queue/max_sectors_kb echo 256 > /sys/block/sdc/queue/max_sectors_kb echo 256 > /sys/block/sdd/queue/max_sectors_kb echo 256 > /sys/block/sde/queue/max_sectors_kb blockdev --setra 64 /dev/sd[bcde] blockdev --setra 16384 /dev/md0 **NOTE:** The files within **/etc/profile.d/** are automatically run at boot, which will result in these tuning options being set at each boot. ---- ===== References ===== https://h3x.no/2011/07/09/tuning-ubuntu-mdadm-raid56 https://ubuntuforums.org/showthread.php?t=1494846 https://ubuntuforums.org/showthread.php?p=11646960 https://askubuntu.com/questions/20852/making-stripe-cache-size-permanent