switch.rst

  1=========
  2dm-switch
  3=========
  4
  5The device-mapper switch target creates a device that supports an
  6arbitrary mapping of fixed-size regions of I/O across a fixed set of
  7paths.  The path used for any specific region can be switched
  8dynamically by sending the target a message.
  9
 10It maps I/O to underlying block devices efficiently when there is a large
 11number of fixed-sized address regions but there is no simple pattern
 12that would allow for a compact representation of the mapping such as
 13dm-stripe.
 14
 15Background
 16----------
 17
 18Dell EqualLogic and some other iSCSI storage arrays use a distributed
 19frameless architecture.  In this architecture, the storage group
 20consists of a number of distinct storage arrays ("members") each having
 21independent controllers, disk storage and network adapters.  When a LUN
 22is created it is spread across multiple members.  The details of the
 23spreading are hidden from initiators connected to this storage system.
 24The storage group exposes a single target discovery portal, no matter
 25how many members are being used.  When iSCSI sessions are created, each
 26session is connected to an eth port on a single member.  Data to a LUN
 27can be sent on any iSCSI session, and if the blocks being accessed are
 28stored on another member the I/O will be forwarded as required.  This
 29forwarding is invisible to the initiator.  The storage layout is also
 30dynamic, and the blocks stored on disk may be moved from member to
 31member as needed to balance the load.
 32
 33This architecture simplifies the management and configuration of both
 34the storage group and initiators.  In a multipathing configuration, it
 35is possible to set up multiple iSCSI sessions to use multiple network
 36interfaces on both the host and target to take advantage of the
 37increased network bandwidth.  An initiator could use a simple round
 38robin algorithm to send I/O across all paths and let the storage array
 39members forward it as necessary, but there is a performance advantage to
 40sending data directly to the correct member.
 41
 42A device-mapper table already lets you map different regions of a
 43device onto different targets.  However in this architecture the LUN is
 44spread with an address region size on the order of 10s of MBs, which
 45means the resulting table could have more than a million entries and
 46consume far too much memory.
 47
 48Using this device-mapper switch target we can now build a two-layer
 49device hierarchy:
 50
 51    Upper Tier - Determine which array member the I/O should be sent to.
 52    Lower Tier - Load balance amongst paths to a particular member.
 53
 54The lower tier consists of a single dm multipath device for each member.
 55Each of these multipath devices contains the set of paths directly to
 56the array member in one priority group, and leverages existing path
 57selectors to load balance amongst these paths.  We also build a
 58non-preferred priority group containing paths to other array members for
 59failover reasons.
 60
 61The upper tier consists of a single dm-switch device.  This device uses
 62a bitmap to look up the location of the I/O and choose the appropriate
 63lower tier device to route the I/O.  By using a bitmap we are able to
 64use 4 bits for each address range in a 16 member group (which is very
 65large for us).  This is a much denser representation than the dm table
 66b-tree can achieve.
 67
 68Construction Parameters
 69=======================
 70
 71    <num_paths> <region_size> <num_optional_args> [<optional_args>...] [<dev_path> <offset>]+
 72	<num_paths>
 73	    The number of paths across which to distribute the I/O.
 74
 75	<region_size>
 76	    The number of 512-byte sectors in a region. Each region can be redirected
 77	    to any of the available paths.
 78
 79	<num_optional_args>
 80	    The number of optional arguments. Currently, no optional arguments
 81	    are supported and so this must be zero.
 82
 83	<dev_path>
 84	    The block device that represents a specific path to the device.
 85
 86	<offset>
 87	    The offset of the start of data on the specific <dev_path> (in units
 88	    of 512-byte sectors). This number is added to the sector number when
 89	    forwarding the request to the specific path. Typically it is zero.
 90
 91Messages
 92========
 93
 94set_region_mappings <index>:<path_nr> [<index>]:<path_nr> [<index>]:<path_nr>...
 95
 96Modify the region table by specifying which regions are redirected to
 97which paths.
 98
 99<index>
100    The region number (region size was specified in constructor parameters).
101    If index is omitted, the next region (previous index + 1) is used.
102    Expressed in hexadecimal (WITHOUT any prefix like 0x).
103
104<path_nr>
105    The path number in the range 0 ... (<num_paths> - 1).
106    Expressed in hexadecimal (WITHOUT any prefix like 0x).
107
108R<n>,<m>
109    This parameter allows repetitive patterns to be loaded quickly. <n> and <m>
110    are hexadecimal numbers. The last <n> mappings are repeated in the next <m>
111    slots.
112
113Status
114======
115
116No status line is reported.
117
118Example
119=======
120
121Assume that you have volumes vg1/switch0 vg1/switch1 vg1/switch2 with
122the same size.
123
124Create a switch device with 64kB region size::
125
126    dmsetup create switch --table "0 `blockdev --getsz /dev/vg1/switch0`
127	switch 3 128 0 /dev/vg1/switch0 0 /dev/vg1/switch1 0 /dev/vg1/switch2 0"
128
129Set mappings for the first 7 entries to point to devices switch0, switch1,
130switch2, switch0, switch1, switch2, switch1::
131
132    dmsetup message switch 0 set_region_mappings 0:0 :1 :2 :0 :1 :2 :1
133
134Set repetitive mapping. This command::
135
136    dmsetup message switch 0 set_region_mappings 1000:1 :2 R2,10
137
138is equivalent to::
139
140    dmsetup message switch 0 set_region_mappings 1000:1 :2 :1 :2 :1 :2 :1 :2 \
141	:1 :2 :1 :2 :1 :2 :1 :2 :1 :2