Author: Martin Czekalski, Maxtor Corporation
Server blades first appeared about two years ago as a way to deliver increased server compute capabilities within a small footprint. Server blades are offered in a wide range of performance and packaging options. High-density blades (Figure 1) offer modest performance with a focus on low power and high packaging density to maximize the number of servers per rack. This high-density form factor has typically used mobile drives for low power. High performance blades (Figure 2) are focused on providing the maximum compute power in a given space, hence they typically use higher performance enterprise drives (10K and 15K) in a 3.5″ form factor. Both blade types utilize a direct-attached storage model for these embedded drives. Additional storage is typically provided by some form of fabric-attached storage (NAS or SAN).
High Density Blade
High Performance Blade
While fabric-attached storage offers flexibility, it comes at the cost of complexity in both the management and boot process. The direct-attached option provides a simplified model for boot and management. The introduction of Serial Attached SCSI (SAS) coupled with increasing capacities, new form factors and low-cost RAID on Motherboard (ROMB), will provide a wide range of architectural choices for designers of blade servers and systems. These choices will allow a wide range of price, performance and packaging options to be offered.
Let’s take a look at some of these options. Figure 3 below illustrates that SAS can be used to create large configurations of both server blades and storage while maintaining the simplicity of the direct-attached storage model. In this example, SAS expanders are used to connect a large number of server blades to one or more RAID subsystems. By doing this, the disk drives located directly on the blades are optional. Removing the disk drives from the blade creates a utility model for the compute environment where the system is divided into compute resources and storage resources, each of which can be managed and optimized independently.
With the advent of inexpensive ROMB controllers for SAS, another configuration option is RAID on the server blade. This allows the total system RAID performance to scale proportionally to the number of server blades in the system. This configuration is shown in Figure 4 below. Additionally, this approach eliminates the need for separate RAID subsystems and allows the end user to “pay as you grow”-reducing initial investment, while still providing expandability. In this example, the blade server chassis still contains the expander function, but the enclosures need only be JBODs (just a bunch of disks), making them less expensive than a larger RAID subsystem.
There are, of course, many variations on these configurations, but it can be seen that SAS provides a high degree of scalability and performance. Additionally, disk drives will be available in a wide range of price and performance points that will allow better optimization and overall system efficiency. SAS drives will be available in 15K RPM for the highest performance needs and 10K RPM for a balance of cost and performance, while SATA drives will be available in 7.2K RPM for high capacity storage at a modest cost. As SAS systems start to roll out over the next several months, system administrators and users will start to reap the benefits.