Playing with blocks or files

Editor’s Note: Mike Karp is on vacation this week.

In recent articles, Mike Karp has discussed the emergence of content addressed storage as outlined by EMC, and contrasted that to network-attach storage and storage-area network solutions. Today, we examine iSCSI, which is support by Microsoft, Cisco, and others, and how it relates to NAS and SANs.

The foundational question of storage access technology is, “block or file?” In other words, “How do the applications needing the storage best access it, and what are the parameters for the access in terms of performance, sharing, and ownership of the actual storage device?”

The original introduction of file servers and NAS addressed the need for sharing files, and provided access to storage as a virtual file system. In this environment, the servers managed all of the disk management, and the storage was completely abstracted as a native file system for each client.

Meanwhile, mainframe systems and later, SANs, used block-mode access to storage. In this environment, the client systems see the storage resource as raw disk space, and manage the space as though it was an internal, local disk drive. The advantage to this approach is that the central storage pool appears as a local disk, with similar performance characteristics, and is invisible to applications.

While these two approaches to the technology remain today, an interesting disruption occurred with the evolution of TCP/IP into a ubiquitous networking technology. This characteristic has driven down the cost of IP-based networking and provided a high-speed infrastructure at much lower cost than historical block-mode networking technologies like ESCON (Enterprise Systems Connection) and Fibre Channel. The subsequent introduction of iSCSI makes this even more interesting.

Prior to the widespread availability of iSCSI, IP networks were the domain of NAS while SANs used block-mode protocols, primarily Fibre Channel. However, iSCSI is a block-mode protocol running on top of TCP/IP, providing the new possibility of leveraging TCP/IP equipment and expertise to support a standards-based, block-mode SAN.

These technology wars do not actually focus on the fundamental implementation issue, which is: “What is the best technology for this specific application?” Instead of jumping to conclusions based on gut feel, storage architects need to take the time to explore the real impacts of the various options available.

For example, instead of assuming that the databases require block-mode access to storage, learn about the possible advantages of using NAS and file system virtualization for databases. Similarly, instead of assuming that NAS is the only option for sharing the same disk space among a cluster of servers, explore emerging shared file systems like IBM’s General Parallel File System that provide shared disk file systems allowing multiple nodes block-mode access the same files.

In short, take the time to consider that unfamiliar configurations may provide additional advantages – and watch the emerging technologies to see how they can make a difference.