When is a storage network not a storage network? Some SAN specialists quibble that NAS isn’t really storage networking as it is just another box plugged in to the existing IT infrastructure. No separate storage network, they say, has actually been established. And there is some truth to this. Now another vendor is arguing that both Fibre Channel (FC) and IP SANs are not true SANs as they require dedicated hardware whereas Storage over IP (SoIP) does not.
SoIP can be used to access storage devices over TCP/IP networks. As the storage becomes accessible to users directly across the network, much of the overhead imposed by server and operating system intervention is removed.
“Fibre Channel and iSCSI have done a good job of putting storage on a network, but they are not really network storage,” says Ryan Malone, director channel marketing for Zetera Corp, a technology licensing company based in Irvine, CA. “They use IP as an extension cord between a server and a disk array that still uses direct-attach disk commands.”
According to Malone, FC and iSCSI have multiple fabrics – one for the network and one for the disks (RAID). SoIP takes a different approach, using IP as the only fabric. SoIP addresses groups of disk blocks as IP addresses. Using IP commands like unicast, broadcast and multicast, SoIP delivers true virtualization of storage as opposed to what Malone characterizes as the brute force version offered by FC and ISCSI vendors.
SoIP solutions are defined by three critical criteria. First, SoIP uses IP as the only storage fabric so its products enable the direct mapping of IP addresses to each storage element without controllers or aggregators. Secondly, SoIP virtualizes storage elements through IP, which is inherently virtualized. IP’s use as the storage fabric eliminates the need for virtualization software. Thirdly, SoIP leverages the power of IP to scale, stripe, mirror and transfer data without regard to the physical location of the physical drives.
The first commercial realization of SoIP technology has been created by Zetera and Marvell Semiconductor of Santa Clara, CA. They have combined forces to match Zetera’s Z-SAN distributed controller architecture with Marvell’s Orion family of processors. Bell Microproducts in San Jose, CA, plans to be the first distributor to release this technology in its Hammer Storage brand during the first half of 2006. The Hammer line will include desktop and rack-mounted equipment with capacities up to 2 TB. Later this year, Hammer products are slated to achieve a capacity of 24 TB.
“Zetera’s Z-SAN technology enables products that simultaneously optimize performance, scalability, reliability and cost,” says Malone. “It offers the speed and sophistication of Fibre Channel with the simplicity of a NAS while scaling up to 128 petabytes per volume without virtualization software.”
With other types of storage, says Malone, customers must choose between various factors. FC, for example, is fast and scalable but at the expense of high cost. On the other hand, iSCSI offers lower performance and scalability but at reduced price. Meanwhile, NAS is low cost, but again runs into barriers in terms of scalability and performance.
Zetera is attempting to position the Z-SAN as filling the niche between FC and iSCSI-based SANs. The protocol enables disk storage to be connected to an IP network and accessed as an IP-addressable element. Groups of IP-addressable elements can be combined into large, high-performance volumes that live anywhere on the IP network. It is accessed by the OS as a local drive letter (i.e., C Drive) and, says Malone, offers block-level performance exceeding Fibre Channel, along with file and volume sharing typical of a NAS with excellent scalability.
“Already, there is a lot of interest in Z-SAN technology, particularly from companies in surveillance, medical imaging, document management, disk-based backup and video post-production,” he says.
Z-SAN technology is enabled by two principal components: The first is a client driver that interprets file system commands and converts them to block-level network transmissions; and, the second component is a network adapter (or logical controller) that receives commands and converts them to disk-oriented I/O commands/payloads.
This approach enables the technology to offer several capabilities: disaggregating physical disks into virtual IP-addressable partitions; aggregating IP partitions into spanned, striped, mirrored and parity-protected logical volumes; the features and performance of RAID hardware without its costs; full file and volume sharing provided on a block-level device without sacrificing performance or manageability; and unlimited scalability via the distributed controller architecture.
Users can deploy Z-SAN to create RAID constructs of up to 128 petabytes per logical volume without the use of virtualization software and without regard to the physical location of the physical disks therein.
“Products based on Z-SAN technology have a lower manufacturing cost than any other network storage architecture,” says Malone. “They eliminate the need for dedicated switches, RAID controllers, HBAs, TOEs and NAS heads.”
This article was first published on EnterpriseITPlanet.com.