The Basics of SAN Implementation, Part II
Most of the attention on SANs has focused on the performance benefits of a dedicated gigabit network that relieves conventional LANs of data movement loads. But, from a more holistic perspective, SANs will provide other significant advantages such as improved storage implementation, manageability, more reliable and flexible backup operations, and shared storage resources among multiple servers.
For peak demand periods, SAN-based implementations offer the ability to allocate additional resources to priority applications and servers. While server re-allocation is possible without SANs, such an approach is far less useful as storage resources cannot be shared. A powerful combination is dynamic server allocation with the ability to add or change storage resources without pre-determination.
One of the most attractive features of SAN implementation technology is its impact on standard network operations. The heavy overhead that conventional storage architectures place on LANs and network file servers is eliminated by relocating storage resources to an independent network.
With the preceding in mind, this article continues the SAN implementation theme presented in Part I, by briefly discussing other SAN implementation topics with regards to backups, clusters, appliances and database applications. Let's look at backups first.
SAN Backup Implementation
Backup operations, typically CPU intensive processes, will be completely removed from the servers. Faster, more reliable backup operations are a key component of SAN implementation. Indeed, the first generation of significant SAN-based applications will be built around a new generation of backup technologies such as:
- LAN-Free Backup.
- Server-Free Backup.
- Zero Backup Window.
- Multiple Small/Medium Libraries Versus One Large Library.
Enterprise storage resources reside on an independent gigabit-speed network in a SAN implementation. All data movement occurs over this high-speed dedicated network and not a standard Ethernet LAN. The effect of SAN-based, LAN-free backup is an immediate improvement in LAN performance.
LAN-free backup technology gives multiple servers access to a single tape library connected to the SAN. Rather than the conventional Ethernet LAN, all backup operations are now routed through the gigabit-speed Fibre Channel SAN.
A new generation of SAN-aware backup software supports this architecture. In order to share tape libraries and eliminate data movement over the LAN, the backup software, being SAN-aware, coordinates between servers to allocate tape library resources. The first widely deployed storage management application to emerge on Storage Area Networks is likely to be LAN-free backup.
Server-free backup takes the LAN-free backup concept a step further. Not only are all backup operations relocated from the LAN to the SAN, but by enabling direct data movement between SAN devices, SAN bandwidth is maximized. In the case of backup, this means that the data moves directly from Redundant Array of Independent Disks (RAID) disk storage to the tape library; and, thus removes the server bottleneck.
Server-free backup leverages two key technical developments: the small computer system interface-3 (SCSI-3) block copy command (also known as third-party copy) and the Network Data Management Protocol (NDMP)-compliant software to manage communications between the server and the tape library. Also, for this application, the term "server-free" backup is actually somewhat of a misnomer. By using the NDMP to manage communications between the SAN storage devices, the server still plays a role in the backup operation and ensures that the backups are completed successfully. However, as in traditional backup operations, server intervention is minimized and all data is sent directly over the SAN and not via the server. This significantly increases performance, while improving the reliability of automated backup processes. As data moves from a server-attached RAID device to a local or network-attached tape library, conventional LAN backup operations consume a vast amount of resources, including server CPU cycles, I/O busses and LAN bandwidth. All aspects of server and network operations are impacted, thus imposing a significant performance hit.
As data is transferred over the high-speed Fibre Channel SAN directly between the source and target storage devices, server-free backup architectures remove virtually all of this processing overhead, thus eliminating traditional backup overhead. Server-free backup solves the backup dilemma by:
- Leveraging Fibre Channel bandwidth to dramatically increase the rate at which data can be moved.
- Eliminating repeated data movement by enabling direct transfers between SAN storage devices.
- Reducing the server resources required to move the data.
- Delivering this functionality on live production systems.
Zero Backup Window
The combination of NDMP-compliant backup applications and SCSI copy technology has the potential to enable another powerful SAN-based application: zero backup window. A "snapshot" of the backup data is created--in effect a point-in-time virtual mirror that requires only a small fraction of the disk space needed to create an actual mirror of the data. Instead of remaining unavailable for the duration of the backup operation, this allows the applications to be returned to production status almost immediately. The snapshot directs the backup software to the disk location of the original data for backup. If a write command is issued to update the backup data set, it is intercepted and the update is written to a new section on the disk, thus maintaining the integrity of the original data. When the backup is completed, the snapshot is deleted freeing up that disk space.
One Large Library versus Multiple Small/Medium Libraries
While the traditional architecture has been to choose one larger library, a SAN also enables universal access to multiple libraries. The advantage of multiple smaller libraries is redundancy (therefore uptime) and cost savings. Large silos often cost hundreds of thousands of dollars, take up significant floor space and are costly to maintain. And if they fail, you cannot continue to perform a backup. You can actually save space, cost, and never be down by having multiple smaller libraries. Having multiple small/medium libraries also allows for cost-effective redeployment of a number of units outside of the SAN.