By Wayne
Rash
July 26, 2002
THE
BASIC CONCEPTS behind iSCSI (Internet SCSI) aren't new.
Companies have been using data networking
technology as a medium for storage
data for years. What's new is the approach, as well as
the level of flexibility that iSCSI provides compared to
earlier versions of NAS
or SANs.
NAS devices, which include iSCSI devices, are simply
storage devices that include NICs (network interface
cards), which make the devices accessible via a standard
network connection.
In the past, such devices have used proprietary
protocols or have appeared on the network as servers,
allowing network users to connect to them. But the
devices did not appear as available storage on servers.
Furthermore, because they were simply objects on the
network, they suffered network traffic congestion along
with everything else.
iSCSI devices are different in that they are accessed
via an iSCSI HBA (host bus adapter). To the computer,
the HBA looks just like any other SCSI HBA: It appears
to be a storage device that you'd access via the server,
just as you would with a directly attached storage
device, such as an internal disk or a SAN. To the
network, the iSCSI HBA appears to be a NIC. It has an IP
address and communicates using standard IP network
packets.
The difference is, when the server needs to move some
data to storage, it transfers the data to its HBA, where
it becomes standard SCSI-3 data. The data is then
enclosed in an IP packet and is sent out via the
Ethernet network. Once it gets to the iSCSI storage
device, the IP packet information is stripped off, and
the data is moved to the device's internal SCSI
controller, which in turn transfers it to disk.
One advantage of iSCSI is that it's completely
transparent. The server software sees only what appears
to be a SCSI controller; the network only sees IP
traffic. To the IT staff, it means that there's little
new to learn. iSCSI uses standard Ethernet
infrastructure and standard SCSI provisions in the
server software.
iSCSI's simplicity makes this new storage protocol
highly flexible. For example, your storage network can
share infrastructure with your standard data network,
just as you might with other NAS devices.
Because servers are talking to the network through their
HBAs, implementing a separate storage network is easy
and relatively inexpensive. Furthermore, because you're
talking to your storage through Ethernet rather than
Fibre Channel, you gain even greater flexibility.
Consider distance, for example. A standard SAN using
Fibre Channel is limited to speeds of 2Gbps and fiber
lengths of approximately 30 kilometers. In testing at
our lab at the University of Hawaii, we were able to
extendFibre Channel communications to about 40
kilometers under controlled conditions. This is handy if
you want to have off-site mirroring, but suppose your
MAN (metropolitan area network) extends farther than 40
kilometers?
Likewise, there's the speed issue. Fibre Channel can
currently handle speeds as fast as 2Gbps. Not bad for
handling storage for an in-house SAN. But suppose you
need to aggregate several SANs for that off-site
mirroring?
In both cases, iSCSI provides a solution in 10GbE
(10-Gigabit Ethernet). This new Ethernet specification
is five times as fast as the fastest SAN, and it covers
vastly more distance. Even better, the infrastructure
for 10GbE is already available, so iSCSI at 10Gbps is
also available.
Continued
>>
|
