IBM N Series, Руководство по оборудованию

Страница: 210 / 368

190
IBM System Storage N series Hardware Guide
The minimum acceptable practice is to allocate one spare drive, per drive type, per
controller head. In our example, this results in two disks because it is a two-node cluster.
򐂰
RAID: When a drive fails, it is the following RAID information that allows the lost data to be
recovered:
– RAID-4: Protects against a single disk failure in any RAID group and requires that one
disk is reserved for RAID parity information (not user data).
Because disk capacities increased greatly over time with a corresponding increase in
the risk of an error during the RAID rebuild, do not use RAID-4 for production use.
The remaining 11 drives (per controller) are divided into 2 x RAID-4 groups and require
two disks to be reserved for RAID-4 parity, per controller.
– RAID-DP: Protects against a double disk failure in any RAID group and requires that
two disks be reserved for RAID parity information (not user data).
With the IBM System Storage N series, the maximum protection against loss is
provided by using the RAID-DP facility. RAID-DP has many thousands of times better
availability than traditional RAID-4 (or RAID-5), often for little or no more capacity.
The remaining 11 drives (per controller) that are allocated to 1 x RAID-DP group
require two disks to be reserved for RAID-DP parity, per controller.
򐂰 The RAID groups are combined to create storage aggregates that then have volumes
(also called
file systems
) or LUNs allocated on them.
Normal practice is to treat the nine remaining disks (per controller) as data disks, which
creates a single large aggregate on each controller.
The following 24 available disks are now allocated:
򐂰
Spare disk drive: 2 (1 per controller)
򐂰
RAID parity disks: 2 (2 per controller)
򐂰
Data disks: 18 (9 per controller)
About 25% of the raw capacity is used by hardware protection. This amount varies depending
on the ratio of data disks to protection disks. The remaining usable capacity becomes less
deterministic from this point because of ever increasing numbers of variables, but a few firm
guidelines are still available.
Right-sizing
A commonly misunderstood memory requirement is that imposed by the right-sizing process.
This overhead is the result of the following main factors:
򐂰 Block leveling
Disks from different batches (or vendors) can contain a slightly different number of
addressable blocks. Therefore, the N series controller assigns a common maximum
capacity across all drives of the same basic type. For example, this process makes all
“1 TB” disks exactly equal.
Block leveling has a negligible memory requirement because disks of the same type are
already similar.
򐂰 Decimal to binary conversion
Because disk vendors measure capacity in decimal units and array vendors usually work
in binary units, the stated usable capacity differs.
However, no capacity is lost because both measurements refer to the same number of
bytes. For example, 1000 GB decimal = 1000000000000 bytes = 931 GB binary.