
                              Hewlett-Packard AutoRAID
                             (The rules have changed...)

New storage technology, HP AutoRAID, automates RAID level selection, optimizes
disk storage capacity and performance and makes fault-tolerant subsystems
easier to configure and manage.

Users of today's RAID subsystems have to pay a performance and administration
price for protecting their data.  In on-line transaction processing
environments, current RAID subsystems have poor performance and high costs
when compared to non fault-tolerant disk subsystems.  With traditional
fault-tolerant subsystems, the performance varies with the configured RAID
level.  Selecting the appropriate RAID level usually involves cost,
performance and availability tradeoffs.  As an example, RAID 0/1 offers best
overall performance but is a more expensive format than RAID 5 or RAID 3.
The system workload also has an effect on RAID level selection.  Write
operations may be more efficient in one RAID mode than another, or random reads
may be more efficient than random writes within the same RAID level.
HP AutoRAID "understands" these differences in the RAID level performance and
dynamically adapts its algorithms to best meet the demand of the host system.

Highlights:

* Subsystem administration without expertise in RAID technology

* Configurations of non-homogeneous disks within the array- flexibility
  for today's and tomorrow's disks capacities

* Add usable capacity, or change performance characteristics in seconds,
  not hours

* Hot spare disks that are in use to increase performance yet provides the
  same function as traditional hot spare disks

* Dynamically adapts its algorithms to best meet the demand of the
  system workload.

AutoRAID Technology
-------------------

HP AutoRAID technology is a set of new algorithms within the subsystem
controller that manage data block addresses.  Each block has a unique
address on the disk.  After a host computer writes a block of data to the
disk at a specific address, it expects to receive the same data back when the
block is read from that address (I certainly like this feature myself ;-).
Traditional disk arrays use static, pre-defined algorithms to translate the
host block address to a location on the disk.  These translation algorithms
are part of the fundamental definition of RAID levels, such as RAID 3 or
RAID 5.

HP AutoRAID introduces the ability to intelligently translate or map, any host
block address to any disk drive address with a dynamic algorithm, and to
change the translation while the system is operating.  This allows the
controller to move data stored within the subsystem to any location on any
disk drive without affecting the data, or how the host addresses the data.

Dynamic Data Migration
----------------------

HP AutoRAID provides two key benefits to the system administrator.  First,
it eliminates the need for the administrator to select or understand RAID
levels and their effect on performance and availability.  Second, the subsystem
continuously optimizes for cost and performance, as if the administrator were
spending all their time tuning the subsystem.

The block address translation facility is the key to HP AutoRAID technology.
The subsystem controller can convert data stored in one RAID level to another
RAID level without disrupting the host view of the data.  The dynamic data
migration strategy is to keep the active data in RAID 1, which offers best
overall performance, and the less active data, which is lower in cost.
Typically active data is only a small portion of the total data at any given
time period.  Thus the majority of the data is stored in RAID 5.  This
strategy provides the performance of RAID 1 with the lower cost of RAID 5.

Log Structured RAID 5
---------------------

HP AutoRAID technology introduces log structured RAID 5 algorithms to
minicomputer and PC systems.  Currently available on some very expensive
high-end mainframe computers storage subsystems, log structured increases
RAID 5's performance.  Traditional RAID 5, utilizing block striping,
sequentially accessed host blocks are written to different disks within the
array.  An XOR (exclusive OR) of all the data within a stripe is sent to
another disks to provide redundant information.  Traditional RAID 5 is very
inefficient in maintaining the XOR data in a random write workload.  RAID 5
requires a read-modify-write, which is four separate I/O's, to complete one
random write.  Log structured RAID 5 is also block striped, with the
exception that each block is not required to be from a sequential address.
Thus, when HP AutoRAID is migrating data from RAID 1 to RAID 5, the data can
be gathered from many random addresses to be written in one complete strip.
These large writes get RAID 3 performance, which delivers the best RAID level
for large transfers.

Mixed Disks and On-line Expansion (now this is cool...)
-------------------------------------------------------

Another feature of HP AutoRAID technology is the ability to easily mix and
match disks of different capacity or performance.  Although some traditional
disk arrays have a similar feature, their configuration process is complex and
time consuming.  With HP AutoRAID the process is simple and fast.  Like
traditional disk arrays, one of the configuration tasks of the administrator
is to create virtual disks from the physical storage space available.  The
array controller then presents the virtual disks to the host computer as if
they were real physical disks.  A traditional disk array requires the
administrator to understand the characteristics of each physical disk drives
to manually group the drives or sections of the drives together to build a
virtual disk.

With HP AutoRAID, the administrator is simply informed of the total space
available on all the disk drives.  The administrator then specifies how much
of this available space is to be used by each virtual disk.  The subsystem
mapping policies then group the disks together to guarantee the largest
usable capacity and the best performance while assuring the the data is
protected by RAID technology.

HP AutoRAID greatly simplifies the addition of capacity.  To add more storage
capacity to a subsystem, the administrator simply installs the disks and
builds another virtual disk.  As with the initial configuration process, the
administrator is informed of the total available capacity including the new
disks.  This process is done on-line and takes only seconds to complete.



Tuning for Performance
----------------------

Unique to this storage technology is the ability to automatically and
immediately use new disks to increase subsystem performance.  When a new disk
is installed, HP AutoRAID will redistribute the data evenly across all of
the disks in the subsystem.  This process, called balancing, is done in the
background between host operations.   With the data distributed evenly across
all disks in the subsystem there are more opportunities for concurrent
(simultaneous) operations.  In transaction processing environments, the more
concurrent operations, the better the performance.

Active Hot Spare  (even cooler...)
----------------------------------

Another innovation of HP AutoRAID is the active hot spare.  Active hot spare
provides the same functions as a hot spare in traditional disk arrays.  That
is, when a disk fails, the controller can immediately start the rebuild
process without operator intervention.  The rebuild process will reconstruct
the missing data and restore the redundancy of the subsystem.  In traditional
disk arrays, the hot spare remains inactive until a failure.  With HP
AutoRAID, the hot spare is used to improve performance of the subsystem.  This
innovation is done by balancing the user data across all the disks in the
subsystem, including the hot spare (better performance), and by reserving a
portion of space on each disk for the reconstructed rebuild data.  When a
disk failure occurs, the rebuild process will reconstruct the missing data
and store it in the reserved space on each disk.


This is as much information as we have available now.  Stay tuned for
future updates on HP AutoRAID.

-----------------------------------------------------------------------------


Phil Haseltine
Response Center Disk Consultant
Atlanta

+---------------------------------------------------------------------------+
| Phil Haseltine - Atlanta Response Center /    Hewlett-Packard Company     |
|---------------------------------------------------------------------------|
|     Email: prh@atl.hp.com              / If it's broke, we fix it.        |
| Home mail: philh@mindspring.com       / If it's not broke, we enhance it. |
|     Phone: (404) 648-0000            / If you fine it acceptable, your    |
|       Fax: (404) 850-2715           / expecations are probably too low.   |
|---------------------------------------------------------------------------|
|          HOME PAGE:  http://www.mindspring.com/~philh/home.html           |
|---------------------------------------------------------------------------|
|            My opinions are mine and whoever agrees with them.             |
+---------------------------------------------------------------------------+


