HP NetRAID Product Paper

Performance
Many factors influence the performance of a
server disk subsystem. They include the total
number of disks and the maximum data transfer
rate of each drive. In a RAID environment, the
ability of the controller to efficiently manage the
data organization is also a major contributor.
Up to six HP NetRAID Adapters can be
connected to a server, with three channels on
each DAC. All three channels are Ultra-wide SCSI
channels with up to six Ultra-wide SCSI drives
per channel, maximizing storage for each
controller and allowing for as much expansion as
necessary.
The HP NetRAID Adapter includes an intelligent
ASIC that handles data transfer between the PCI
bus, cache, and SCSI bus, and performs RAID
parity generation and checking for seven RAID
levels. Because software-based RAID uses host
CPU cycles and system memory, it adds overhead
that may impact system performance. The HP
NetRAID Adapter moves the burden of RAID
computations and manipulation from software to
specialized hardware and often performs better
than software RAID.
Several enhancements provided by the HP
NetRAID Adapter improve read/write
performance. An adaptive read-ahead algorithm
improves read performance. A write-back with
cache memory feature improves write operations
in RAID-3 or RAID-5. These features, as well as
turning cache on or off, can be changed
interactively for real-time performance tuning.
RAID levels 10, 30, and 50 (described in the next
section) combine striping and parity to balance
read/write performance.
Flexibility and Data Protection
The HP NetRAID Adapter provides the highest
degree of flexibility in data storage and
protection with seven RAID levels and such
features as online capacity expansion, RAID level
migration, and drive roaming.
RAID Levels The HP NetRAID Adapter supports
four industry-standard RAID levels: striping
(RAID-0), mirroring (RAID-1), duplexing (RAID-1
with dual controllers), striping with dedicated
parity (RAID-3), and striping with distributed
parity (RAID-5). The NetRAID Adapter supports
three additional levels—RAID-10, RAID-30, and
RAID-50—that combine levels 1, 3, and 5 with
RAID-0, increasing I/O throughput and handling
larger volume sizes. (This paper does not explain
each RAID level in detail. For more information
on RAID levels, please refer to HP’s “Technology
Brief on RAID Levels.” All HP technology briefs
and white papers are available on the HP Web
site at http://www.hp.com/go/netserver.)
This range of choices in RAID levels gives you
the ability to select a disk availability solution to
meet your needs for performance, recovery time,
flexibility, and cost. As you might expect,
increasing levels of redundancy typically come at
higher costs, so it is important to match the level
needed to the data and applications being used.
Online Capacity Expansion Online capacity
expansion makes it possible to add a physical
drive to a logical drive while the server is in
operation, eliminating the need to store and
restore data after reconfiguring the array. Online
capacity expansion increases storage flexibility
and makes future growth easy to accommodate.
By increasing the size of the logical drive,
capacity expansion provides better management
and data integrity than other expansion methods,
such as adding a separate array and logical drive.
Online capacity expansion is currently available
with Microsoft Windows NT and Novell NetWare.
Banyan VINES, IBM OS/2, and SCO UNIX require
rebooting the server to recognize the new
capacity.
RAID Level Migration Another tool for managing
your array solution is RAID level migration. RAID
level migration enables changing RAID levels on
the fly as server requirements change. For
example, in a system using two drives in RAID-1
(mirroring), the system administrator could add
capacity and retain fault tolerance by adding one
drive and migrating to RAID-5. The result would
be parity fault tolerance and double the available
capacity without taking the server down. Or, as
another example, if a system using RAID-5 with
three drives experienced a drive failure and had
no hot spare or available spare drive, the system
administrator could migrate to RAID-0 and
maintain high performance until a new drive
became available. Again, the server stays up and
is not running RAID-5 with degraded
performance. Once the new drive is available, the
array can migrate back to RAID-5.