Demystifying Storage Networking
Demystifying Storage Networking
Demystifying Storage Networking
June 2001
by David Sacks
IBM Storage Consultant
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 2
1
Midrange is essentially shorthand for non-mainframe, non-standalone PC.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 3
In a Nutshell.
Well start with a brief description of the major storage networking variations.
The paper will then develop the concepts in a more structured manner.
SANergy: SANergy is
software from IBM and
Tivoli that provides NAS-
like file sharing, with data
sent over the SAN rather
than the LAN for improved
performance. (IBM NAS
gateways also include
SANergy function.)
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 5
Why are there so many forms of storage networking? For one, new
technologies emerge and evolve but dont replace the investment in previous
technologies overnight. And no single storage networking approach solves
all problems or optimizes all variables. There are tradeoffs in cost, ease-of-
management, performance, distance and maturity, to name a few of these
variables. For the foreseeable future, multiple storage network alternatives
will coexist often within the same organization.
The benefits of the major types of processor-to-storage connectivity can be
briefly summarized as:
DAS is optimized for single, isolated processors and low initial cost.
SAN is optimized for performance and scalability. Some of the major potential
benefits include support for high-speed Fibre Channel media which is optimized
for storage traffic, managing multiple disk and tape devices as a shared pool
with a single point of control, specialized backup facilities that can reduce
server and LAN utilization and wide industry support.
NAS gateways are optimized to provide NAS benefits with more flexibility
in selecting the disk storage than offered by a conventional NAS device.
Gateways can also protect and enhance the value of installed disk systems.
Tivoli SANergy is optimized for data sharing (like a NAS), but at SAN speeds.
Tivoli SANergy is disk vendor-independent, and can be added to an existing
SAN to enhance its value.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 6
Connectivity.
The pictures below illustrate the two basic ways to physically connect storage
to processors.
The simplest form of direct attached storage (DAS) is a single disk drive or
single tape drive connected to a single processor. Some disk systems allow
the aggregate disk capacity to be carved into partitions (subsets) of capacity
where each partition can be assigned to a different processor. Further, the
subsystem may allow partitions to be manually reassigned from one processor
to another.2 This is essentially still a DAS approach to storage.
Direct attach can be thought of as a minimal network. For simplicity,
and as is common in the industry, this paper will sometimes refer to storage
networking alternatives without explicitly mentioning direct attach, but it
should be considered as one such alternative.
Following industry convention, a cloud is used to indicate a network
without showing the inner details of how cables, and devices such as hubs
and switches, may be connected to form a particular implementation. Such
implementations will vary from organization to organization and do not
need to be understood in order to explain storage connectivity alternatives.
The idea is that all objects connected to the same cloud can potentially
communicate with each other. (Such any-to-any flexibility can be managed
in practice to prevent undesired communications.)
2
For example, the IBM Enterprise Storage Server disk system offers this exibility.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 8
Media.
The media is the physical wiring and cabling that connects storage
and processors.
Media is always managed by a low-level protocol unique to that media
regardless of the attached devices. A protocol is the rules for exchanging
information between two objects. In computers, this specifies the format
and sequence of electronic messages. In storage-to-processor connections,
the following media and associated protocols are prominent. All are open,
industry standards.
Ethernet: Ethernet began as a media for building LANs in the 1980s. Typical
bandwidths are 10Mbps, 100Mbps, and 1Gbps.3 Ethernet is a media and its
protocol. IP-based protocols such as TCP/IP generally run on top of Ethernet.
Fibre Channel: Fibre Channel is a technology developed in the 1990s that
has become increasingly popular as a storage-to-processor media (for both
SANs and DAS). Bandwidth is generally 100MBps, with 200MBps expected
in 2001.
Parallel SCSI (Small Computer Systems Interface): (Pronounced
scuzzy). Parallel SCSI is an evolving technology with origins in the
1980s. Typical bandwidths are 40MBps (also called UltraSCSI), 80MBps
(also called Ultra2 SCSI), and 160MBps (also called Ultra160 SCSI). Parallel
SCSI is limited to relatively short distances (25 meters or less, maximum)
and so is appropriate for direct attach, especially when storage and processors
are in the same cabinet, but is not well-suited for networking.
SSA (Serial Storage Architecture): SSA is a media technology optimized
for high-performance and used to connect disks together inside some disk
systems. Bandwidth is 160MBps.
3
MBps=megabytes/second, Mbps=megabits/second, and Gbps=gigabits/second. 1Gbps generally
equals 100MBps since the (Ethernet and Fibre Channel) protocols involved use special 10-bit bytes.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 9
I/O Protocols.
I/O processing uses specific protocols that run on top of the underlying
media protocols. (In the case of Ethernet, I/O protocols generally run at
some level on an IP protocol stack.) The following are the most common
I/O protocols supported on midrange platforms.
With SCSI (block-I/O), disk volumes are visible to the servers attached to
them. With NFS and CIFS (file-I/O), only files are visible to the attached
processors, but the disk volumes on which those files reside are not visible
to those processors.4
4
While similar in principle, NFS and CIFS differ in many aspects such as user authorization and locking
protocols. For the purposes of this guide those differences are unimportant. Other protocols that deal with
les but not di sk vol ume s incl ude FTP (Fi le Tr ansf er Pr ot ocol ) for transmi tting ent ire les over anet wor k,
and HTTP (Hypertext Transfer Protocol) for transmitting Web pages over a network. These protocols are
not further discussed in this paper, though they are supported by some NAS appliances.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 10
5
For example, it is possible to run TCP/IP over Fibre Channel and so use Fibre Channel as a LAN, and thus
potentially use it for NFS and CIFS requests. However, this is rarely if ever implemented in practice.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 11
Note that while the terms NAS and SAN seem similar, SAN refers to a
dedicated storage network and NAS is a device on a LAN/WAN network
(whether the network is shared or dedicated to storage). Occasionally, the
industry uses the term SAS to refer to SAN Attached Storage. As you may
realize, storage networking terminology is not intuitive, and isnt standardized;
you may want to take care that you and others are talking about the same
thing when using a given term.
A Tabular Comparison.
The various storage networking alternatives are summarized in the following
table.
A Tabular Comparison
Tivoli SANergy Yes SAN media NFS, CIFS, SAN speeds Yes Yes
SCSI
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 12
Legend.
6
In general, every operating system, including every UNIX-based variant, stores data in a format that only
that same operating system understands. File-I/O puts data on the network so that operating systems can
access it using industry-standard protocols without any dependence on data format.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 13
Protection: LAN-free backups occur over the SAN rather than the (slower)
LAN, and server-free backups can let disk storage write itself directly to
tape without processor overhead.
Data Sharing: sharing data, as noted earlier, offers benefits such as
reducing the number of copies of files, increasing accessibility to current
data and reducing the need to transfer copies of data between servers over
the network.
Ease-of-installation.
NAS is generally easier to install and manage than a SAN. A NAS appliance can
usually be installed on an existing LAN/WAN network. NAS manufacturers often
cite up and running times of 30 minutes or less. (Customization procedures
may take additional time.) Hosts can potentially start to access NAS storage
quickly, without needing disk volume definitions or special device drivers. In
contrast, SANs take more planning, including design of a Fibre Channel network
and selection/installation of SAN management software.
Backup.
Most NAS appliances in the marketplace include a snapshot backup facility, to
make backup copies of data onto tape while minimizing application downtime.
For SANs, such facilities are available on selected disk systems or in selected
storage management packages.
Resource pooling.
NAS allows capacity within the appliance to be pooled. That is, the NAS
device is configured as one or more file systems, each residing on a specified
set of disk volumes. All users accessing the same file system are assigned space
within it on demand. That is certainly more efficient than buying each user
their own disk volumes (DAS), which often leads to some users having too
much capacity and others too little. So NAS pooling can minimize the need to
manually reassign capacity among users. However, NAS pooling resides within
a NAS appliance, and there is little if any sharing of resources across multiple
appliances. This raises costs and management complexity as the number of
NAS nodes increases. In contrast, an advantage of a SAN is that all devices on
a SAN can be pooled multiple disk and tape systems. So, at some point as
total capacity grows, a SAN may be easier to manage and more cost effective.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 18
File sharing.
NAS provides file sharing, but with products like SANergy discussed later, a
SAN can do this as well. Many organizations install a NAS, not for file sharing,
but for its ease of installation and management.
Performance.
How do NAS and SAN performance compare? It may depend on the
particular configuration, but SAN is generally considered to be faster. This
is mainly due to:
For relatively low amounts of activity, NAS and SAN may both perform
acceptably well. Today, however, NAS will generally not scale as well as SAN
in performance. It is not clear where the break even point is, but NAS
devices often can handle several thousand I/Os per second with good average
response time (e.g., under 10 milliseconds average for small random I/Os).
To summarize the comparison between NAS and SAN, while a NAS
appliance is generally less scalable and less grandiose than a SAN, it can
satisfy storage requirements in numerous environments ranging from small
businesses to workgroups or departments in large organizations. NAS alone
is, and will remain, a good fit in many environments. NAS and SAN hybrids
(by way of NAS gateways, discussed below) will be a good fit in the largest
environments, combining the best of both worlds.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 19
NAS will generally cost more than DAS (because of its built-in file sharing
intelligence), but has the following potential advantages: distance (because it is
attached over a network), large number of users being able to access the same
storage device, capacity pooling within the NAS appliance (sharing capacity
among all hosts using the NAS), and file sharing (as opposed to data transfer
or multiple copies on distributed hosts).
NAS appliances support standard file access protocols such as NFS,
CIFS, and sometimes others, that run over an IP network. These protocols
were developed before dedicated NAS appliances existed, and are often
implemented in software that runs on most client and server processors. So, in
fact, anyone could build their own NAS device by taking a server of any size
and installing NFS programming on it, for example. NFS is actually supported
directly by most operating systems, or is available from software vendors. The
builder or integrator can use any disk products they want, even a single,
internal disk for a small NAS built using a low-cost desktop PC.
Building your own NAS means flexibility. But buying an integrated NAS
means less time, assurance that the package works, vendor support for the
package, and usually specialized software tuned for the NAS environment and
thus providing much higher performance than possible in a general purpose
server and OS environment.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 20
NAS Gateways.
SANergy.
In brief, SANergy is
software from IBM
and Tivoli that provides
NAS-like file sharing, with
data sent over the SAN
rather than the LAN for
improved performance.
Some in the industry
are calling SANergy and
similar facilities SAFS -
SAN Attached File
Systems.
SANergy has attributes of NAS and SAN, with additional flexibility.
SANergy supports the
NFS and CIFS protocols, but allows the installation to use virtually any disk
storage they want (Fibre Channel, iSCSI, parallel SCSI, and SSA storage will
all work.)
Here is a typical SANergy scenario. A set of processors run SANergy
client software. The initial CIFS or NFS request for a file is intercepted by
the SANergy client and sent over a LAN to a processor running SANergy
Meta Data Controller (MDC) software which handles standard CIFS and NFS
protocol functions such as authorization. The SANergy client dynamically
transmits the actual I/O (data) traffic over the LAN or over the SAN,
whichever is optimal.
Functionally, SANergy supports the protocols of a conventional NAS
appliance but with significantly higher performance while not requiring the
dedicated NAS processor front-end to the disk storage. Instead, SANergy sits
as software in the client hosts (plus the MDC). See www.tivoli.com/sanergy
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 22
Consider the following scenario that illustrates how IBM NAS gateways and
SANergy can work together:
A Web server receives HTTP requests for Web pages and sends them to an
IBM NAS gateway which in turn connects to disks over a SAN. Performance
is degraded due to a large volume of Web pages being returned to the server
over the LAN. So, the installation adds an adapter connecting the server to
the SAN, adds SANergy client software to the Web server, and enables the
SANergy MDC in the gateway. Now, Web pages travel from the disk to the
Web server directly at SAN speeds. If traffic increases so that high server
utilization becomes the bottleneck, then a second server with a SANergy
client could be added, and connected to the MDC and the SAN similar to
the first server. Both servers access the same Web pages at high-speed by
using SANergy.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 23
iSCSI.
Because the concepts and products surrounding DAS, SAN and NAS
preceded iSCSI, it is natural to try to understand where iSCSI fits in the
world by comparing it to those concepts.
The fundamental technical difference between iSCSI and NAS is that iSCSI
is block-I/O oriented while NAS is file-I/O oriented. The fundamental
technical difference between iSCSI and Fibre Channel SANs is that iSCSI uses
TCP/IP networks. Therefore, iSCSI devices fill a void by uniquely supporting
block-I/O applications over TCP/IP (usually Ethernet) networks.
The small table below summarizes this discussion. The columns show
media alternatives, while the rows show how Block I/O and File I/O are
supported on the media.
Future Directions.
The storage networking industry is moving so fast that any predictions should
be treated cautiously. Certainly, higher speed media, both 200MBps Fibre
Channel and faster Ethernet are expected soon.
The ability for organizations to implement open SANs and mix-and-match
heterogeneous vendor storage and network components is increasing as experience
with storage networks grows and as standards for interoperability evolve and are
complied with. Tivolis Storage Network Manager, for example, is a vendor-neutral
SAN management product that adheres to open industry standards.
The industry is developing specialized chips and device adapters that
will offload the TCP/IP protocol handling from the host and disk system
processor, making iSCSI (and probably NAS as well) increasingly practical
in more I/O-intensive environments. While iSCSI will likely start small, it is
expected to increase in capability and popularity over time, providing SAN
benefits such as scalability and storage network-oriented management tools,
but without the need for a specialized Fibre Channel network.
NAS, SAN, and iSCSI will be increasingly converging. For example, if a
NAS appliance is on a LAN dedicated to just the NAS storage traffic, it is
SAN-like in its dedication to storage. A NAS gateway appears NAS-like to
clients, but may attach to disks or tape through a backend Fibre Channel SAN.
With iSCSI, a SAN can be built using Ethernet media, which is the media NAS
generally uses today. Organizations will have increasing ability to customize
storage connectivity to their particular needs, but the choices also mean more
expertise is needed to make the best decisions.
Demystifying DAS, SAN, NAS, NAS Gateways, Fibre Channel, and iSCSI.
Page 28
The table below identifies a few simple scenarios and perspectives on what may
be effective storage connectivity approaches.
An organization has only a very small Either NAS, updated direct-attached storage,
number of servers and low I/O loads, or iSCSI are likely best. A Fibre Channel SAN
but wants to replace installed, aging may not be justi able or necessary. Compared
direct-attach disk storage. to direct-attach and iSCSI, NAS offers better
sharing of capacity even if there is no le
sharing, and simpler management, but it will
likely cost more than DAS. If disk system
functions like snap backup are of value,
that may tip the scale in favor of NAS.
An organization has an existing SAN SANergy preserves the SAN and adds le
using a variety of disk systems and sharing. Or, a NAS gateway could be placed
wants to do some le sharing. in front of the disk system. Or, the les to be
shared could be moved to a NAS, of oading
some SAN traf cif that is of value.
A large organization has heavy I/O A SAN will likely provide the best performance.
loads including heavy database
activity against a relatively small
amount of capacity.
An organization needs a large amount A NAS Gateway allows multiple users to access
storage for a temporary project but an existing SAN for available storage, without
does not have access to a SAN. requiring direct access to the SAN (e.g.,
without installing Fibre Channel adapters on
each host). After the project completes, the
storage can be released back to the SAN for
use by other users. Snapshot backup functions
are also available through the gateway. An
alternative would be to add iSCSI or NAS to
an existing LAN, and later redeploy its capacity
to other projects.
An organization wants to use its LAN iSCSI supports this, allowing I/Os to ow over a
for disk storage but has applications LAN without the need to install a SCSI or Fibre
that use SCSI block-I/O protocols. Channel Host Bus Adapter in the servers.
Summary.
This paper has explored the exciting area of storage networks. If it has
clarified what can be a rather complex subject, then it has been a success.
For more information
Please contact your IBM marketing
representative or an IBM Business
Partner. For more information about
IBM Storage Solutions, visit:
storage/ibm.com.