Many kinds of SCSI protocols have evolved over the years: SCSI-1, SCSI-2, SCSI-3, Ultra 2 SCSI, Ultra 3 SCSI, and Serial Attached SCSI just to name a few. SCSI protocols are standards that exist to define how data is transferred from the data initiator (server) to the storage target (disk array). Until iSCSI, which stands for the Internet Small Computer Systems Interface, came along, it was not possible to transfer SCSI data where the distance between the storage initiator and storage target was further than 25 meters.
With the creation of Storage Area Networks, people wanted to be able to have their storage targets farther away from the data initiators. Due to the distance limitations of SCSI, new connectivity mechanisms arose such as fibre channel. In the last six years, many new fibre channel devices have evolved, including host adapters, fibre channel fabric switches, and device interfaces. Fibre channel connections can span distances of 10 kilometers, and, if you connect one fibre channel switch to another and use special optic transceivers, you can use fibre channel over distances of up to 100 kilometers.
While this may seem like a long way, it is still a limitation which some organizations may find unacceptable. That’s where iSCSI fits in. With iSCSI, you can send your storage blocks across any distance that can be accommodated by a network. From a storage perspective, this represents the ultimate in flexibility in so far as where storage devices can be placed in relation to each other. The ramifications of iSCSI are enormous, and its use stands to revolutionize the way businesses run and manage their storage operations. In essence, anywhere an IP network can go, iSCSI can go. Certainly the business continuity and disaster recovery potential for iSCSI is enormous. With your storage systems in distant locations that are removed from your operations, there is no danger that your data will be lost in the event of a catastrophic event.
Using standard Ethernet networks with iSCSI to move the data from the server to the storage target, there exists the capability to initiate the data in, say, New York, and store it in California. Since consolidated storage serves multiple servers writing to single disk arrays, economies of scale can be gained through using this type of architecture. For example, traditionally, data backups have occurred through file transfers on LANs. With iSCSI, data backups can occur at the block level, which can greatly accelerate the backup process, free up network resources, and make it possible to backup to a distant remote device.
Though iSCSI holds a lot of connectivity promise, and offers new distance capabilities that were previously not possible, certainly the Fibre Channel and InfiniBand vendors would not want you to think so. Fibre Channel vendors will tell you that iSCSI is an inherently slow transport mechanism, and they are likely right about that. However, there is a price to pay for the performance gains. Fibre Channel, and its even newer connectivity foe InfiniBand, are both more expensive to implement than iSCSI.
InfiniBand is a nascent connectivity solution, and not many organizations are ready to become early adopters in its storage and connectivity solutions at this point. Though InfiniBand may emerge as the ideal connectivity solution for diskless clusters in the future, it will likely not impact either the Fibre Channel or iSCSI markets for many years.
Other technologies such as Fibre Channel over IP (FCIP), and Internet Fibre Channel Protocol (iFCP) stand to offer the same distance capabilities as iSCSI, without the potential throughput and latency concerns that are associated with iSCSI. However, it should be noted that iSCSI is a client/server connection oriented protocol, whereas FCIP is a gateway-to-gateway protocol. FCIP is an encapsulation protocol used for transporting Fibre Channel frames over TCP/IP networks. FCIP’s main purpose is to connect Fibre Channel SANS to IP networks thereby extending the distance capabilities of pure fibre channel SANs.
While iSCSI, FCIP, and iFCP all appear to be similar technologies, their ideal implementation scenarios are very different. None of these connectivity protocols offer any inherent security, and will all require a security protocol such as IPSec to protect the data they are transmitting.
Naturally the use of iSCSI provokes some questions and concerns. For example, if you need to retrieve the data from a data warehouse farm in California, and you are in New York, what are the performance ramifications and what will the latency be upon executing an application?
Another major consideration is that of security. Is running iSCSI over a VPN the only safe way to transport storage data? What sort of latency and performance issues will using iSCSI over an IPSec based VPN create? On a more subtle level, but still of concern, is whether third-party data farms will accommodate iSCSI, and who will be the third-party iSCSI players providing this service?
Notwithstanding these concerns, all of which are addressable in some form or another, iSCSI holds a lot of promise, but there will be performance challenges in executing large-scale implementations. That said, the implementation costs of iSCSI are low enough that it will likely appeal to IT shops that may not have the budget for Fibre Channel.