ng

US 2003/0131182 A1
METHODS AND APPARATUS FOR
IMPLEMENTING VIRTUALIZATION OF
STORAGE WITHIN A STORAGE AREA
NETWORK THROUGH A VIRTUAL ENCLOSURE
BACKGROUND OF THE INVENTION
[0001] 1. Field of the Invention
[0002] The present invention relates to network technol
ogy. More particularly, the present invention relates to
methods and apparatus for supporting virtualiZation of stor
age Within a storage area netWork.
[0003] 2. Description of the Related Art
[0004] In recent years, the capacity of storage devices has
not increased as fast as the demand for storage. Therefore a
given server or other host must access multiple, physically
distinct storage nodes (typically disks). In order to solve
these storage limitations, the storage area netWork (SAN)
Was developed. Generally, a storage area netWork is a
high-speed special-purpose netWork that interconnects dif
ferent data storage devices and associated data hosts on
behalf of a larger netWork of users. HoWever, although a
SAN enables a storage device to be con?gured for use by
various netWork devices and/or entities Within a netWork,
data storage needs are often dynamic rather than static.
[0005] FIG. 1A illustrates an exemplary conventional
storage area netWork. More speci?cally, Within a storage
area netWork 102, it is possible to couple a set of hosts (e.g.,
servers or Workstations) 104, 106, 108 to a pool of storage
devices (e.g., disks). In SCSI parlance, the hosts may be
vieWed as “initiators” and the storage devices may be
vieWed as “targets.” Astorage pool may be implemented, for
example, through a set of storage arrays or disk arrays 110,
112, 114. Each disk array 110, 112, 114 further corresponds
to a set of disks. In this example, ?rst disk array 110
corresponds to disks 116, 118, second disk array 112 corre
sponds to disk 120, and third disk array 114 corresponds to
disks 122, 124. Rather than enabling all hosts 104-108 to
access all disks 116-124, it is desirable to enable the
dynamic and invisible allocation of storage (e.g., disks) to
each of the hosts 104-108 via the disk arrays 110, 112, 114.
In other Words, physical memory (e.g., physical disks) may
be allocated through the concept of virtual memory (e.g.,
virtual disks). This alloWs one to connect heterogeneous
initiators to a distributed, heterogeneous set of targets (stor
age pool) in a manner enabling the dynamic and transparent
allocation of storage.
[0006] The concept of virtual memory has traditionally
been used to enable physical memory to be virtualiZed
through the translation betWeen physical addresses in physi
cal memory and virtual addresses in virtual memory.
Recently, the concept of “virtualiZation” has been imple
mented in storage area netWorks through various mecha
nisms. VirtualiZation interconverts physical storage and vir
tual storage on a storage netWork. The hosts (initiators) see
virtual disks as targets. The virtual disks represent available
physical storage in a de?ned but someWhat ?exible manner.
VirtualiZation provides hosts With a representation of avail
able physical storage that is not constrained by certain
physical arrangements/allocation of the storage.
[0007] One early technique, Redundant Array of Indepen
dent Disks (RAID), provides some limited features of vir
Jul. 10, 2003
tualiZation. Various RAID subtypes have been implemented.
In RAID1, a virtual disk may correspond to tWo physical
disks 116, 118 Which both store the same data (or otherWise
support recovery of the same data), thereby enabling redun
dancy to be supported Within a storage area netWork. In
RAIDO, a single virtual disk is striped across multiple
physical disks. Some other types of virtualiZation include
concatenation, sparing, etc. Some aspects of virtualiZation
have recently been achieved through implementing the
virtualiZation function in various locations Within the stor
age area netWork. Three such locations have gained some
level of acceptance: virtualiZation in the hosts (e.g., 104
108), virtualiZation in the disk arrays or storage arrays (e.g.,
110-114), and virtualiZation in a storage appliance 126
separate from the hosts and storage pool. Unfortunately,
each of these implementation schemes has undesirable per
formance limitations.
[0008] VirtualiZation in the storage array is one of the
most common storage virtualiZation solutions in use today.
Through this approach, virtual volumes are created over the
storage space of a speci?c storage subsystem (e.g., disk
array). Creating virtual volumes at the storage subsystem
level provides host independence, since virtualiZation of the
storage pool is invisible to the hosts. In addition, virtualiZa
tion at the storage system level enables optimiZation of
memory access and therefore high performance. HoWever,
such a virtualiZation scheme typically Will alloW a uniform
management structure only for a homogenous storage envi
ronment and even then only With limited ?exibility. Further,
since virtualiZation is performed at the storage subsystem
level, the physical-virtual limitations set at the storage
subsystem level are imposed on all hosts in the storage area
netWork. Moreover, each storage subsystem (or disk array)
is managed independently. VirtualiZation at the storage level
therefore rarely alloWs a virtual volume to span over mul
tiple storage subsystems (e.g., disk arrays), thus limiting the
scalability of the storage-based approach.
[0009] When virtualiZation is implemented on each host,
it is possible to span multiple storage subsystems (e.g., disk
arrays). A host-based approach has an additional advantage,
in that a limitation on one host does not impact the operation
of other hosts in a storage area netWork. HoWever, virtual
iZation at the host-level requires the existence of a softWare
layer running on each host (e.g., server) that implements the
virtualiZation function. Running this softWare therefore
impacts the performance of the hosts running this softWare.
Another key dif?culty With this method is that it assumes a
prior partitioning of the available storage to the various
hosts. Since such partitioning is supported at the host-level
and the virtualiZation function of each host is performed
independently of the other hosts in the storage area netWork,
it is dif?cult to coordinate storage access across the hosts.
The host-based approach therefore fails to provide an
adequate level of security. Due to this security limitation, it
is difficult to implement a variety of redundancy schemes
such as RAID Which require the “locking” of memory
during read and Write operations. In addition, When mirror
ing is performed, the host must replicate the data multiple
times, increasing its input-output and CPU load, and increas
ing the traf?c over the SAN.
[0010] VirtualiZation in a storage area netWork appliance
placed betWeen the hosts and the storage solves some of the
dif?culties of the host-based and storage-based approaches.