#171; back to results for ""
Below is a cache of http://www.syncsort.com/PDF/10530-678.pdf. It's a snapshot of the page taken as our search engine crawled the Web.
The web site itself may have changed. You can check the current page or check for previous versions at the Internet Archive. Yahoo! is not affiliated with the authors of this page or responsible for its content.
Backup strategies for networked storage However, this middle group is flanked by
extremes: On one side are those who view
storage area networks (SANs) as too costly,
complex, and time-intensive to maintain, and
on the other side are those who see network-
attached storage (NAS) as a low-end product
that does not effectively meet the needs of
many storage-intensive applications.
However, a middle ground is coming, and
there are many new technologies and tools in
the pipeline that will aid in meeting the con-
siderable challenge of backing up data on het-
erogeneous networks, whether SAN, NAS, or
a hybrid of both. Two major areas of advance-
ments are in interconnect technologies and
evolving protocols from the Storage
Networking Industry Association (SNIA).
Network-attached storage
NAS optimizes the Ethernet, or Gigabit
Ethernet, IP networking technology of a LAN.
A NAS device (or
server) is a box
with a minimal ker-
nel operating sys-
tem, designed to
facilitate data move-
ment. It is platform-
agnostic, and any user using
virtually any operating system
can send data to or receive data from a NAS device using file-oriented
protocols such as NFS, CIFS, or HTTP over TCP/IP. Clients access stor-
age on a NAS server on a file-by-file basis, as opposed to accessing stor-
age block-by-block over Fibre Channel or iSCSI. Attached to the LAN,
a NAS server has an any-to-any connection, enabling shared and remote
storage over a heterogeneous network.
The Network Data Management Protocol (NDMP) is used to send
backup instructions over TCP/IP to a NAS device. Backup becomes a
much easier taskand more standardizedwith the NDMP protocol.
NDMP will be discussed in more detail later in this article.
Storage area network
A SAN is a network built specifically to handle storage and backup traf-
fic. It is separate and distinct from the LAN and takes storage traffic off
the regular network. For all practical purposes, at this point all SANs use
Fibre Channel interconnects, although technologies such as iSCSI will
become more viable next year.
In a Fibre Channel SAN, every server and storage device is linked
directly to the SAN, so it is a many-to-many connection on a dedicated
storage network. Any of the servers on the SAN can access any of the
storage devices, allowing sharing of expensive tape libraries and other
storage hardware between multiple servers.
Because it uses Fibre Channel, hosts and applications see storage
devices attached to the SAN as locally attached storage. The Fibre
Channel architecture uses block I/O protocols, and so it is higher speed
than TCP/IP-based protocols. Because of its network characteristics, a
SAN can support multiple protocols and operating systems and can be
managed much the same way as a heterogeneous network is managed. It
BY S O U B I R A C H A R YA
A N D S U S A N G . F R I E D M A N
Backup strategies for
networked storage
A quick review of NAS, SAN, iSCSI, serverless backup, NDMP,
snapshots, and DAFS in the context of backup.
R
ecently there has been a plethora of arti-
cles on SANs, NAS, and NAS vs. SAN
for network storagealmost an overload of
opinions and controversy. The prevailing wis-
dom is that eventually they will not be seen as
separate and dichotomous solutions. Instead,
they will be viewed as complementary aspects
to an overall network storage strategy.
Reprinted with revisions to format, from the November 2001 edition of InfoStor
Copyright 2001 by PennWell Corporation is also relatively scalable, after overcom-
ing the first large-scale building step.
iSCSI
iSCSI is an emerging storage networking
technology that is currently in the stan-
dards approval process in the Internet
Engineering Task Force (IETF). Among
other potential advantages, iSCSI is tai-
lor-made for heterogeneous networks.
iSCSI is a TCP/IP-based storage net-
working specification that enables any
storage connected to an IP network to
be backed up from any point on that
network. By carrying SCSI commands
over an IP network, iSCSI facilitates
long-distance storage deployment,
management, and data transfer over
intranets. This capability makes
deployment flexible. Storage and
servers can be added at a convenient
location, not in locations dictated by
proximity, as with Fibre Channel. Since
its sent over TCP/IP, iSCSI can sustain
multiple simultaneous channels, allow-
ing many-to-many conversations
between hosts and storage.
Because IP (Ethernet and Gigabit
Ethernet) networks are pervasive in all
areas of networkingthe Internet,
WANs, and LANsthey can make use
of iSCSI. Storage traffic can be carried
over the existing intranet/Internet with-
out any need to upgrade. There is already
considerable knowledge and investment
in TCP/IP networking, making it very
likely that end users will adopt a protocol
that makes use of the existing infrastruc-
ture. In this sense, iSCSI is a threat to
Fibre Channel SANs, which require
investing in new and separate plumbing.
iSCSI maps a serial SCSI protocol on
top of TCP to enable a host computer to
see and move data in and out of remote
IP-attached storage devices as if they
were local devices. Disk or tape storage
plugged directly into a LAN is then visi-
ble to all computers in the LAN, much
like any regular computer or server
would be. This is ideal for backup con-
siderations because it means that, subject
to security restrictions, any client can
back up to any device, regardless of its
location.
i
SCSI or NAS?
Both iSCSI and NAS provide the flexi-
bility of multiple, shared storage
devices over existing TCP/IP networks,
but each has advantages and disadvan-
tages. A NAS device is essentially an
appliance with a power cord and net-
work cable that comes plug-and-play
ready. Extensive hardware configura-
tion is not required.
iSCSI will similarly come with plug-
and-play ease. However, creating an
iSCSI disk device requires a lower level
of investment for vendors, compared to
creating a NAS appliance, which
requires an operating system, file sys-
tem, and other components. All that is
needed for an iSCSI disk device to work
is a network card, and associated logic,
that can act as an iSCSI target, extract-
ing SCSI commands from TCP packets.
An iSCSI disk subsystem is a better
solution than directly (locally) attached
storage because there is less disruption
when adding additional storage. Storage-
on-demand becomes easy when there is
no downtime: Just add another device to
the network. Conversely, if storage is
directly attached, downtime is almost
guaranteed when additional units are
connected.
iSCSI disk subsystems will deliver
data to hosts faster than a NAS arrange-
ment, because the protocol is block I/O-
oriented and does not have the overhead
of a separate network protocol to talk to
a client. In the early stages, we will also
see iSCSI gateways that are plugged into
a gigabit network at one end, with SCSI
or Fibre Channel storage behind it. This
could provide an easy migration path,
leading to early adoption.
iSCSI is still in its infancy, however,
and the existing network infrastructure
and TCP/IP protocol are not well-suited
for storage, which requires high band-
width and relatively error-free transmis-
sion. Traditional SCSI implementations
have assumed a dedicated link to storage
with very low error rates and low-laten-
cy burst transfers. TCP/IP has to be
adapted to meet these requirements.
iSCSI implementations must employ a
gigabit network as well as modified
TCP/IP stack software for optimal per-
formance. Some kind of storage virtual-
ization software will also be needed as
an enabling technology to orchestrate all
the pieces involved in this storage pool-
ing.
In addition, iSCSIs newness as a pro-
tocol introduces interoperability issues.
Competing vendors are developing dif-
ferent implementations that may not
work together. Much the same problem
is found with Fibre Channel SANs.
Emerging industry standards generated
from the IETF will hopefully address
these issues, rendering them moot in the
near future.
Quality of Service requirements may
need to be enforced to achieve a work-
able storage network. iSCSI has require-
ments such as the ability for Point A and
Point B to negotiate dedicated band-
width for a period of time, which net-
works do not normally provide. Network
security can also be an issue. However,
the iSCSI protocol negates that concern
because it incorporates peer-to-peer
authentication as well as encryption, if
necessary. An iSCSI login phase, during
which session characteristics are negoti-
ated, precedes actual data transfer.
Serverless