Sun Storage 9970 System
 
Sun Storage 9990 System
 
Sun Storage 9960 System
 
Sun Storage 9980 System
 

GCS>Sun Microsystems>Storage - Disk>Datacenter Disk
 

There is a significant difference between linear and dispersed addressing.  However, both types of addressing are completely equivalent from a performance standpoint, as long as the administrator takes the type of addressing into account when configuring volumes.

Before discussing the two types of addressing, it will be helpful to define another key concept:  the parity group. A parity group is a logically related group of four or eight hard disk drives (HDDs).  Parity groups are physically arranged in the array so that they contain no single points of failure.  For example, each member of a parity group is on a separate fibre loop, and can be accessed by two separate disk adapters (DKAs), two separate fibre switches (FSWs), etc.  Typically, parity groups are further subdivided into logical devices (LDEVs) which are presented to hosts as logical units (LUNs).  When data is written to one of the LDEVs on a parity group, each of the HDDs in the group receive an equal portion of the data.1

LDEVs are identified with a control unit (CU) and logical device number in the format CU:LDEV.  When linear addressing is used, the CU and LDEV numbers are assigned sequentially within a single parity group (PG), until the space in that PG is exhausted.  For example, parity group 1-1 might contain LDEVs 00:00 through 00:1F.  PG 1-2 might continue with LDEVs 00:20-00:3F, and so forth.  This type of addressing is perfect for the administrator who likes to keep track of which parity groups correspond to which logical devices.  When the administrator needs to obtain maximum performance by striping I/O across multiple parity groups and DKAs, it is relatively easy to precisely identify which parity groups to use. Also, when capacity is added, it is much easier to distinguish the  new  LDEVs from the ones that were already on the array.  Figure 1 shows an example of an array that has been set up with linear addressing.  Note that the sequentially numbered LDEVs are all on the same parity group.

The primary disadvantage of linear addressing is that it requires attention to the physical location of  LDEVs on parity groups location when configuring volumes.  Otherwise, the administrator may configure the volume manager such that a great deal of I/O is focussed on only one or two parity groups.  For example, if the administrator wants to configure a 4-way striped volume and selects four consecutively numbered LDEVs to make up the columns, the entire volume would reside on a single parity group. If the customer's workload requires a high number of I/Os per second (IOPS), a concentration on only one or two PGs will not perform well.  See <Document: 1018062.1> for more information.

Dispersed addressing allocates sequentially numbered LDEV addresses to different parity groups.  For example, LDEV 00:00 would be placed on parity group 1-1, and LDEV 00:01 would go on PG 1-2, and so on.  Dispersed addressing is most appropriate for the administrator who prefers not to track the physical location of LUNs. Unlike the situation with linear addressing, if the administrator chooses four sequentially numbered LDEVs to compose a striped volume, these LDEVs will be on four separate parity groups, and the I/O will be well-dispersed. The primary disadvantage of dispersed addressing is that it makes more difficult to map LDEVs to parity groups. This is particularly true if the array has had capacity added several times, as the addressing scheme tends to become more confusing and jumbled with each upgrade.  If performance problems occur, or if other issues arise that require determining what the physical distribution of LDEVs is, the required mapping will be more difficult with dispersed addressing.  Figure 2 shows an example of dispersed addressing.

[ 1 ] In RAID5, one of the drives will receive parity data instead of customer data, and the drive receiving parity will be changed for the next write. In RAID6, two copies of parity data are written.

This solution has no attachment