HP StorageWorks Virtual Array 7000 Family User and Service Guide (January 2005)

Product Overview 49
Product Overview
simultaneous disk failures. The two redundancy segments are referred to as
“P” and “Q” parity. P, like traditional RAID 5 arrays, uses an XOR (parity)
algorithm. P parity is based on Reed-Solomon ECC technology, similar to
error detection and correction found in ECC DRAM.
Application data, and the P and Q parity data, rotate to different disks for
each stripe in a RAID Group. Like RAID 1+0, this effectively eliminates hot
spots.
A read operation only requires a single access to the disk(s) containing the
data, a small (<256 Kbytes) write operation requires that the data, and the P
and Q parity data be updated – this is the source of the small random write
performance impact. For larger (>256 Kbytes) write operations, the Virtual
Array implements a log-structured RAID 5DP write. Log-structured writes
effectively eliminate the read-modify-write associated with small block writes to
RAID 5DP by redirected the write operation to a new RAID 5DP stripe. The P
and Q parity data is held in non-volatile write cache until the whole stripe is
written, then the P and Q are written. Thus the P and the Q are written only
once for each stripe.
Note Until a rebuild is complete, the array is operating in a degraded
mode. In degraded mode, the array will use P and/or Q parity
to reconstruct data that resided on the failed disk.
Figure 19 is an example showing the distribution of user data and parity data
in a RAID 5DP configuration. The example shows one RAID group with five
stripes: three data segments and two parity segments (P and Q). The
segments are striped across the disks in a rotating fashion. Note that any two
disks can fail, but the data, P, or the Q parity is always available to complete a
read operation.