Using nvSRAMs for power-fail reliability in enterprise SSDs

SSD Technology Overview

Solid State Drives (SSDs) are data storage devices that use a solid-state semiconductor memory, such as NAND Flash, to persistently store information, instead of a magnetic element as used in traditional Hard Disk Drives (HDDs). This results in a faster input/output (I/O) performance for SSDs since data can be randomly accessed and does not depend on read/write head synchronizing with a rotating disk as is required with HDDs. In addition, the time required to move the head to the correct position in HDDs is in the order of several milliseconds.

The basic architecture of a SSD is composed of a SSD controller/processor, a memory controller, an interface controller, a bank of NAND flash memory devices, SDRAM cache, and an interface connector.

SSDs have no moving parts and emulate HDDs because they are manufactured in the same form factors and support standard HDD interfaces, such as serial advanced technology attachment (SATA), Serial Attached SCSI (SAS), and Fiber Channel (FC). No moving parts result in higher reliability over a longer operational life.

Another major advantage of SSDs is significantly lower power consumption compared to HDDs. As memory capacities increase and prices drop, SSDs are becoming an increasingly attractive alternative to HDDs. Because they are faster, SSDs cost much less per IOPS (input / output operations per second) than HDDs. SSDs are also becoming more cost-effective over time in terms of cost per gigabyte (GB). Analysts expect that SSD prices will continue to fall steadily, spurring increasing adoption of the technology in market segments

Enterprise SSD

Enterprise-grade SSDs represent the highest tier of nonvolatile storage available today and a step-change improvement for storage technology in terms of read/write performance, heat dissipation and energy consumption over HDD alternatives. The enterprise applications that can derive the greatest benefits from SSDs, which act as storage network accelerators, include banking and financial applications, online transaction processing, front-end Web servers, search engines, messaging, and high performance computing.

Because Enterprise SSDs are plug-compatible with HDDs and support standard disk interfaces, they can be installed in most server platforms and disk arrays currently using Enterprise HDDs. The main performance metric for an enterprise-grade storage device is random read or writes IOPS (See Table 1).

 

Table 1: Source: http://en.wikipedia.org/wiki/IOPS

Enterprise SSDs are offered in moderate to high capacity, with strong performance and reliability specifications. They are aimed exclusively at the enterprise storage markets for application acceleration.

Figure 1 shows the basic block diagram of a SATA Interface Enterprise SSD. Other interfaces available are HDD-compatible Serial Attached SCSI (SAS), Fiber Channel (FC), and PCIe.

 

Figure 1 : Enterprise SSD Basic Block Diagram

The following sections discuss the need for SDRAM cache in Enterprise SSDs and the current architecture of using a super capacitor or a bank of tantalum capacitors to back-up the critical portion of SDRAM cache data on power down, as shown in Figure 1. The reliability issues with this implementation are discussed and the use of nonvolatile memory solution (nvSRAM) as a superior alternative is explored.

Need for SDRAM cache

NAND flash memory is the basic storage element in an Enterprise SSD. Due to its architecture, the main limitation of NAND flash memory is that its write speed cannot match the data transfer speeds of Enterprise storage systems. Because data transfer speed exceeds NAND flash write speed, Enterprise SSD write performance can be improved by using a high-speed data cache. Enterprise SSDs typically use SDRAM as a cache to hold and work on portions of the data streams received from the storage system controller. In addition, the SDRAM holds a working copy of the Enterprise SSD metadata, a portion of which must be modified corresponding to allocation of blocks for the data being written. Metadata typically includes information on wear leveling, error correction, translation tables, physical/logical address maps, file allocation tables and so on, and requires multiple write operations for every file. Metadata requirements grow with Enterprise SSD capacity.