In its simplest form, a NAS system for corporate use consists of a standard x86 server running a modified flavor of Linux or Microsoft WSS (Windows Storage Server). The server can either share an enclosure with disk drives or reside in its own dedicated enclosure and add storage capacity with external modules. Resilient solutions add an additional server for active fail-over.
Unified storage solutions give customers more flexibility by offering a common capacity bucket for file systems and SAN volumes alike, making it easier to reach higher utilization on storage devices. These systems frequently handle SAN via iSCSI, a protocol that is a good complement to the traditional file serving protocols for Linux/Unix and Windows, although some products — such as the FAS line from NetApp — also include Fibre Channel connectivity. On the client side, ubiquitous iSCSI initiator software and Gigabit Ethernet NICs make for inexpensive SAN access.
Even these systems, however, can only scale up to a point. They may not satisfy customers who need to manage large files in big volumes quickly. For example, to work around a file system that can’t grow beyond 16TB, customers may have to break their files across multiple file systems, which increases complexity and vulnerability.
“Namespace aggregation and virtualization create the illusion of a single unified system but don’t solve underlying technical constraints,” Goodwin says.
Reaching out to the cluster
To meet the demand of customers such as Kodak and others, Isilon designed IQ, a clustered storage solution with ambitious goals. IQ is designed to support file systems 20 to 50 times larger and 15 to 20 times faster than most NAS solutions. Its OneFS file system is highly reliable and self-healing, with a symmetric, distributed architecture that makes managing thousands of terabytes easy. In addition, IQ includes diversified cluster nodes that maximize performance or capacity and interact with fast InfiniBand connections on the back end.
It’s worth noting that, although Isilon nodes mount SATA (serial ATA) drives, OneFS doesn’t rely on traditional hardware RAID and can recover quickly from two simultaneous failures. “With a 10-node cluster, we can rebuild a 500GB SATA drive in two and half hours,” Goodwin says. “It could take up to 24 hours on a traditional NAS system.”
Moreover, Isilon IQ 4.0 clusters can grow in capacity to as much as 528TB simply by adding more nodes, a quick operation that doesn’t affect running applications (see “Scalable NAS: Just What the Doctor Ordered,” page 34). Files are striped across all nodes, which translates into faster performance as the cluster grows. It also speeds recovery operations because more boxes will be working to rebuild the missing files.
Isilon’s approach isn’t the only way to handle clustered NAS, however, and it may not be the best way to address every storage scenario. For example, an Isilon system has excellent sequential performance because every node contributes to moving a large file quickly, but random-access performance may not be as good.
The ONStor Bobcat Series NAS Gateway is another kind of clustered solution that takes a completely different approach. “Our architecture excels with random performance,” says Jon Toor, ONStor’s vice president of marketing.