In-Depth

Storage Provisioning with Blade Servers

Direct-attached hard drives in servers are problematic. ISCSI SANs may be the solution.

• By Peter Hunter
• 06/01/2004

Blade computing continues the already-established trend away from monolithic server architectures. When combined with sophisticated software, blade servers are the data center manager’s provisioning dream: completely modular computing in an efficient form factor. Sadly, most blade computing designs borrow a problematic feature from monolithic server architectures: a direct-attached hard drive. There is a much better way: combine blade servers with another hot trend in computing iSCSI-based storage area networks (SANs).

Blade servers and iSCSI SANs are a marriage made in heaven. Together they enable modular computing and modular storage interconnecting on a single networking fabric.

The Blade Server Concept

The rationale behind blade servers (and their value proposition to the data center) are founded on the blade server's decoupling of the core components of CPU and memory from the power supply, cooling units, cabling, and chassis. The result: a system in which the server itself is hot-swappable.

Stripped of its bulkier components, the server is placed onto a small sled (or blade) that can be hot-plugged into a chassis with other blades. The chassis contains the power, cooling, and cabling components, often incorporating a network switch or patch board. The blades themselves act as fully autonomous servers. Blades within a single chassis can run different applications and play independent roles. Often blades are clustered together into highly available configurations driving a single application.

Blades significantly reduce the overall footprint of server computing by offering a very dense form-factor. For example, a single 3U (three unit) blade server shelf can contain as many as 16 servers, each with the same power as a typical 1U “pizza box” server. Some blades offer multi-CPU architectures with large amounts of RAM. These blades have the horsepower to compete with enterprise servers. In addition to their reduced footprint, blades tend to be power efficient, significantly lowering operating costs.

However, the real added value of blade computing is in the provisioning capabilities of the hardware and accompanying software packages. The idea behind blade provisioning is to greatly simplify the deployment of new servers and the expansion of existing servers. For example, if two Web servers are participating in a cluster and additional Web servers are needed, the administrator can simply insert additional blades and configure them as Web servers. This provisioning configuration usually is automatic, following a script defined within the blade provisioning software. The operating system and applications are installed automatically. The security and networking information are automatically configured.

Blade Servers and Storage

The smallest form factor blades use hard drives designed for laptops. These low-power, low-heat, compact hard drives seem perfect for the blade architecture. However, these laptop drives are not suitable for enterprise applications that require storage such as databases, mail servers, and file sharing. These blade servers tend to be limited to networking roles, such as DNS, Web, and firewall applications.

Mid-range and “enterprise” blade designs incorporate more robust storage solutions. These approaches include incorporating onto the blades hot-swap drive bays, SCSI drives, and RAID controllers for data protection. Unfortunately, these designs add bulk, heat, power consumption, and cost. Not every blade application will need the same amount of storage. The result is poor asset utilization and wasted capacity. A better solution is needed.

Blade Servers and SANs

The answer lies within another product of the departure from monolithic server designs: storage area networks. By decoupling the storage from the server, SANs put the storage onto a network, vastly improving storage provisioning, availability, and operations. By combining blade computing with SANs, administrators can provision and scale storage and servers independently.

Two solutions for SANs are available: Fibre Channel (FC) and iSCSI. Some blade designs now come equipped with onboard FC host bus adaptors (HBAs) and integrated FC switches. While offering networked storage, the FC approach sacrifices cost. To put things in perspective, the FC HBA is often more expensive than the rest of the blade components, making it difficult to rationalize a blade’s function for anything other than enterprise storage applications (such as a database). Does a DNS server need a FC HBA for its boot drive? Of course not.

Blades and iSCSI SANs

iSCSI offers the best match for SAN-enabled blade computing. From the beginning, blades have shared one thing: they are designed around Ethernet networks. Typically, each blade is equipped with two onboard gigabit Ethernet ports. These ports are wired through the hot-swap blade-chassis backplane directly into integrated Ethernet switches and patch panels, eliminating the cables from the servers to the switches. Since iSCSI is SCSI over TCP/IP and all blade servers have Ethernet ports, all blades are iSCSI ready. In addition, most blade designs incorporate integrated Ethernet switches into their chassis. All an administrator must do is add an iSCSI disk array.

By migrating blades to an iSCSI SAN, the entire system becomes incredibly resilient and flexible. First, availability immediately is escalated to an enterprise level. iSCSI disk arrays are available with the same enterprise-quality redundancy of FC arrays: hot-swappable dual controllers, redundant power, cooling, network ports, and RAID-protected disks. These highly redundant designs vastly improve upon the reliability of traditional blade disks, even the RAID protected ones. In the event that a blade server fails, it can be swapped out and its volumes mounted by a replacement (or standby) blade server. In addition, the blade volumes are now afforded the features of an enterprise SAN. These often include point-in-time copies (also called snapshots), SAN backup, data transport, and remote replication for disaster tolerance.

Flexible, Instantaneous Blade Provisioning

Blade provisioning is vastly improved by leveraging the volume cloning features of some iSCSI disk arrays. Volume cloning is the capability to instantly replicate a particular volume. Need another blade file server? Simply clone the existing file server’s system disk and point the new blade server at the volume -- no lengthy installation required. Most cloning operations can be automated and integrated into the blade provisioning software, making the entire operation an instantaneous affair.

In summary, combining blade computing with iSCSI SANs extends the modular paradigm. Blades are computers on demand, while iSCSI SANs enable storage on demand. The iSCSI SAN broadens the horizon of blade server applications, beyond simple edge-of-the-network solutions. Database clusters, e-mail server clusters, and print-and-file sharing are just a few of the applications blade servers can tackle when connected to an enterprise iSCSI SAN.