IP storage taking center stage
STORAGE-AREA NETWORKS
 
|
|||
 | |||
|
As storage becomes a more indispensable part of a company's infrastructure, the technology to link storage facilities is only beginning to solidify.
Two types of products are starting to appear: those based on the storage-over-IP (iSCSI) protocols that let storage data run over existing Ethernet networks; and the Fibre Channel-over-IP protocol for bridging geographically separated storage-area networks (SAN). The goal for both technologies is to come up with corresponding methods of sending storage over IP that will work in a variety of networked conditions.
There are a number of specifications related to the transport of data over IP, but these two drafts are predominant and could be official standards by year-end. The draft specifications, which are before the Internet Engineering Task Force, are surrounded by a series of other specifications for management, encapsulation of data, remote booting and variations for configuring IP storage networks.
Prestandard products using iSCSI and Fibre Channel over IP have started to appear. IBM has introduced an IP storage array it calls the TotalStorage IP Storage 200i, which uses iSCSI; and Cisco, Nishan Systems and CNT have announced or are shipping Fibre Channel-over-IP or iSCSI storage switches and routers. EMC, Cisco, Nortel Networks, CNT and Lucent have introduced Fibre Channel-over-IP products that tunnel Fibre Channel in IP for transport over dense wave division multiplexing (DWDM).
These vendors say that software for their iSCSI products will be upgradable when the standard arrives, thus preserving customers' investments. Fibre Channel-over-IP vendors, including EMC, CNT, Nortel and Lucent, admit that some equipment changes may be necessary because of the complexity of tunneling Fibre Channel in IP.
Fibre Channel over IP
Fibre Channel over IP is a method for bridging Fibre Channel data from SANs over IP to link geographically diverse sites. It is meant simply to let isolated SANs share data across a network or distributed company. For instance, SAN A in Brooklyn, N.Y., might replicate data for disaster recovery purposes to SAN B in Queens using DWDM routers from Nortel, Lucent or Cisco. Or, SAN C in Detroit might share data with SAN D in Los Angeles for availability purposes - so that two sets of users could have access to the same data mining information at once.
Ivan Santhumayor, a database administrator for a Wall Street firm, says his company uses Fibre Channel-over-IP technology to make data available for several sites at once and to provide data protection.
"We use EMC's Symmetric Remote Data Facility [SRDF] to mirror all disk writes, including [operating system] and database files, across the network," Santhumayor says.
"There is a slowdown in disk I/O performance, but it's worth the price," he says. "We did successful failover tests and were impressed with the way we could switch over the application in a few minutes to the mirrored SRDF site."
In Fibre Channel over IP, Fibre Channel data is tunneled through the IP network from a SAN on one site to a SAN on the other end of the IP connection. The idea is to join distant SANs so that data can be more available or for disaster recovery purposes.
A variant of Fibre Channel over IP is Internet Fibre Channel Protocol, which takes data from existing Fibre Channel hosts or devices and transports it across the IP network using the same protocol iSCSI uses. Nishan, Nortel, Quantum/ATL and Sun are among the vendors that may implement iFCP products.
Analysts say that enterprise networks that have not yet adopted Fibre Channel over IP are unlikely to, and instead will look for an alternative that preserves their existing investments in training and equipment.
"People that haven't invested in Fibre Channel won't in the future, because of cost, complexity and management," says Steve Duplessie, an analyst with Enterprise Storage Group. "It's not that they don't need the benefits of block SANs, it's simply the cost, complexity and management skills have been greater than the potential reward."
"Fibre Channel SANs are the toys of the data elite and will continue to prosper in those environments. The data center will continue with Fibre Channel. Outside the data center - where Fibre Channel is nonexistent - will be the playground of iSCSI," Duplessie says.
iSCSI works
For the company that hasn't implemented a SAN already, iSCSI is an excellent choice because it uses the existing Ethernet infrastructure that users are familiar with, says Arun Taneja, an analyst with Enterprise Storage Group.
"It already has all the physical and virtual tools for managing an iSCSI network," Taneja says. "They have the physical Ethernet infrastructure and the firewalls and security that go along with it. And they have the management tasks because it is IP."
In IP storage, native SCSI data from hosts or storage devices is encapsulated in Gigabit Ethernet packets for transport across the IP network. For instance, SCSI data is encapsulated by the software driver on a Gigabit Ethernet adapter located in the server into Gigabit Ethernet transport for transfer across the network. When the data reaches the iSCSI router or switch, such as Cisco's SN5420 storage router, the router removes the Gigabit Ethernet "envelope" and gives the SCSI data to the Fibre Channel or SCSI storage array or tape library.
"I want a technology that will allow me to grow my network, to increase storage and reduce complexity," says Eric Goldfarb, CTO for publisher Macmillan USA in Indianapolis.
"We have IP everywhere. When you look at independent storage networking, if you keep the vision of one network, everything is so much simpler," Goldfarb says. "If I can get a technology that allows me to manage my network using the same management tools, the same people, the same wiring and the same hardware, whether its fiber, IP or Gigabit Ethernet, I can get economies of scale."
Related Links
