(by: Cameron Brett, former Manager, Alliances Marketing at PMC-Sierra, Inc.)

Tiered storage can have many meanings. It can mean the time-based movement of data from primary storage to an archive, as in hierarchical storage management (HSM). It can be interpreted as ensuring mission critical data is stored on the most protected storage volumes while less important data is stored on less protected storage. Today, it is commonly referred to as the caching or movement of data to flash-based storage for the fastest access of the most frequently accessed data so systems can perform at their optimal capability.

As seen in Figure 1, SSDs and flash storage are now at the top of the storage pyramid, providing extremely fast data storage and access in common form factors of 2.5″ drive and PCI-cards. SAS HDDs, which led the performance way with 15K RPM, were at the top but are now effectively the middle tier. The good news in this type of storage implementation is that SAS HDDs can be fully utilized with no more “short stroking” required. Short stroking uses a portion of the HDD which provides the highest performance of the drive. Below SAS HDDs, SATA HDDs (with their higher capacity and slower performance) and cloud storage serve as nearline or off-site storage which is sometimes used for backup or archive.

“Tiered Storage in a Flash” was the title of a live demonstration at the SCSI Trade Association (STA) Technology Showcase in May 2012. The goal was to take real world measurements comparing today’s fastest SAS hard disk drive (HDD) storage and complimenting it with SAS solid state disks (SSD). Working with actual enterprise storage providers’ equipment, the results were eye-popping, showing a nearly 20x measured improvement in IOPS performance.

What do you need to implement Tiered Storage?

Creating a tiered storage volume is easier than you think. All the necessary components are shipping today from multiple vendors, so the risk is low and the returns are very high.

In a typical server, you have a main HDD storage volume. In some cases, the HDDs are short-stroked to maximize the IOPS performance of the drive, but they limit the amount of usable storage to about 25% of the overall capacity as seen in Figure 2. For some applications, this is acceptable because performance demands outweigh the cost of the HDDs. The result is a 4x requirement in the number of HDDs needed to reach the desired capacity.

SSD capacity as a portion of the overall storage volume can range from 10% to 100%. Since SSDs are still expensive with respect to HDDs, 10% to 25% of SSD storage is commonly used. With so much innovation and development in flash and SSDs, the cost curve will favor more use of flash storage over the next few years. SSDs use SAS and SATA interfaces, like HDDs. This enables the use of the existing infrastructure in enterprise servers and storage. Other flash storage devices use the PCI Express bus with flash memory on a PCIe add-in card form factor.

RAID controllers are still widely used to protect primary storage in servers. SSDs can be used on the same RAID controllers and will be seen by the system as a standard storage device. Newer RAID controllers have the capability of using SSDs specifically as a read or write cache to the RAIDed HDD volume. In some cases, the caching/tiering capability can be added via a software application.

Great Results with Tiered Storage

In HDD-only configurations, short-stroked drives created the highest performance available. Many drives were required to create the desired 4TB of usable storage capacity. At an estimated cost of $400 per 1TB SAS HDD and 400 IOPs per drive, the performance and cost profile is listed in Figure 3.

Adding two SSDs in the server and removing 12 HDDs, the usable 4TB capacity is the same, since short stroking is no longer needed. The overall storage cost is similar with an estimated cost of the SSDs at $2,000 for a 200GB SAS SSD (see Figure 4). Under the same read/write access profile, the performance of the tiered SSD/HDD configuration skyrockets to nearly 20x the performance of the HDD-only configuration. In the end, the application and usage pattern will determine the true performance benefit, but it is clear that significant improvement will be realized. Additional benefits to this configuration include the use of smaller servers that need to support six devices, rather than 16, and the overall system reliability improves with fewer devices. Power requirement are also reduced.

Tiered Storage in a Flash

Tiered storage can have many meanings. In this case, it means vastly better performance, improved uptime and reduced power requirements. SSDs/flash storage is here to stay and the benefits will be greater as the technology continues to improve over the next few years. How quickly can you do this? You guessed it. You can implement Tiered Storage… in a Flash.