Scott Bauer | 18a5bf2 | 2017-12-18 10:28:08 -0700 | [diff] [blame] | 1 | Introduction |
| 2 | ============ |
| 3 | |
| 4 | The device-mapper "unstriped" target provides a transparent mechanism to |
| 5 | unstripe a device-mapper "striped" target to access the underlying disks |
| 6 | without having to touch the true backing block-device. It can also be |
| 7 | used to unstripe a hardware RAID-0 to access backing disks. |
| 8 | |
| 9 | Parameters: |
| 10 | <number of stripes> <chunk size> <stripe #> <dev_path> <offset> |
| 11 | |
| 12 | <number of stripes> |
| 13 | The number of stripes in the RAID 0. |
| 14 | |
| 15 | <chunk size> |
| 16 | The amount of 512B sectors in the chunk striping. |
| 17 | |
| 18 | <dev_path> |
| 19 | The block device you wish to unstripe. |
| 20 | |
| 21 | <stripe #> |
| 22 | The stripe number within the device that corresponds to physical |
| 23 | drive you wish to unstripe. This must be 0 indexed. |
| 24 | |
| 25 | |
| 26 | Why use this module? |
| 27 | ==================== |
| 28 | |
| 29 | An example of undoing an existing dm-stripe |
| 30 | ------------------------------------------- |
| 31 | |
| 32 | This small bash script will setup 4 loop devices and use the existing |
| 33 | striped target to combine the 4 devices into one. It then will use |
| 34 | the unstriped target ontop of the striped device to access the |
| 35 | individual backing loop devices. We write data to the newly exposed |
| 36 | unstriped devices and verify the data written matches the correct |
| 37 | underlying device on the striped array. |
| 38 | |
| 39 | #!/bin/bash |
| 40 | |
| 41 | MEMBER_SIZE=$((128 * 1024 * 1024)) |
| 42 | NUM=4 |
| 43 | SEQ_END=$((${NUM}-1)) |
| 44 | CHUNK=256 |
| 45 | BS=4096 |
| 46 | |
| 47 | RAID_SIZE=$((${MEMBER_SIZE}*${NUM}/512)) |
| 48 | DM_PARMS="0 ${RAID_SIZE} striped ${NUM} ${CHUNK}" |
| 49 | COUNT=$((${MEMBER_SIZE} / ${BS})) |
| 50 | |
| 51 | for i in $(seq 0 ${SEQ_END}); do |
| 52 | dd if=/dev/zero of=member-${i} bs=${MEMBER_SIZE} count=1 oflag=direct |
| 53 | losetup /dev/loop${i} member-${i} |
| 54 | DM_PARMS+=" /dev/loop${i} 0" |
| 55 | done |
| 56 | |
| 57 | echo $DM_PARMS | dmsetup create raid0 |
| 58 | for i in $(seq 0 ${SEQ_END}); do |
| 59 | echo "0 1 unstriped ${NUM} ${CHUNK} ${i} /dev/mapper/raid0 0" | dmsetup create set-${i} |
| 60 | done; |
| 61 | |
| 62 | for i in $(seq 0 ${SEQ_END}); do |
| 63 | dd if=/dev/urandom of=/dev/mapper/set-${i} bs=${BS} count=${COUNT} oflag=direct |
| 64 | diff /dev/mapper/set-${i} member-${i} |
| 65 | done; |
| 66 | |
| 67 | for i in $(seq 0 ${SEQ_END}); do |
| 68 | dmsetup remove set-${i} |
| 69 | done |
| 70 | |
| 71 | dmsetup remove raid0 |
| 72 | |
| 73 | for i in $(seq 0 ${SEQ_END}); do |
| 74 | losetup -d /dev/loop${i} |
| 75 | rm -f member-${i} |
| 76 | done |
| 77 | |
| 78 | Another example |
| 79 | --------------- |
| 80 | |
| 81 | Intel NVMe drives contain two cores on the physical device. |
| 82 | Each core of the drive has segregated access to its LBA range. |
| 83 | The current LBA model has a RAID 0 128k chunk on each core, resulting |
| 84 | in a 256k stripe across the two cores: |
| 85 | |
| 86 | Core 0: Core 1: |
| 87 | __________ __________ |
| 88 | | LBA 512| | LBA 768| |
| 89 | | LBA 0 | | LBA 256| |
| 90 | ---------- ---------- |
| 91 | |
| 92 | The purpose of this unstriping is to provide better QoS in noisy |
| 93 | neighbor environments. When two partitions are created on the |
| 94 | aggregate drive without this unstriping, reads on one partition |
| 95 | can affect writes on another partition. This is because the partitions |
| 96 | are striped across the two cores. When we unstripe this hardware RAID 0 |
| 97 | and make partitions on each new exposed device the two partitions are now |
| 98 | physically separated. |
| 99 | |
| 100 | With the dm-unstriped target we're able to segregate an fio script that |
| 101 | has read and write jobs that are independent of each other. Compared to |
| 102 | when we run the test on a combined drive with partitions, we were able |
| 103 | to get a 92% reduction in read latency using this device mapper target. |
| 104 | |
| 105 | |
| 106 | Example dmsetup usage |
| 107 | ===================== |
| 108 | |
| 109 | unstriped ontop of Intel NVMe device that has 2 cores |
| 110 | ----------------------------------------------------- |
| 111 | dmsetup create nvmset0 --table '0 512 unstriped 2 256 0 /dev/nvme0n1 0' |
| 112 | dmsetup create nvmset1 --table '0 512 unstriped 2 256 1 /dev/nvme0n1 0' |
| 113 | |
| 114 | There will now be two devices that expose Intel NVMe core 0 and 1 |
| 115 | respectively: |
| 116 | /dev/mapper/nvmset0 |
| 117 | /dev/mapper/nvmset1 |
| 118 | |
| 119 | unstriped ontop of striped with 4 drives using 128K chunk size |
| 120 | -------------------------------------------------------------- |
| 121 | dmsetup create raid_disk0 --table '0 512 unstriped 4 256 0 /dev/mapper/striped 0' |
| 122 | dmsetup create raid_disk1 --table '0 512 unstriped 4 256 1 /dev/mapper/striped 0' |
| 123 | dmsetup create raid_disk2 --table '0 512 unstriped 4 256 2 /dev/mapper/striped 0' |
| 124 | dmsetup create raid_disk3 --table '0 512 unstriped 4 256 3 /dev/mapper/striped 0' |