Flash Disks for Telecom Storage

With the declining price of flash, replacing the mechanical HDD in telecom networks with a flash disk has become an attractive alternative.

By Esther Spanjer

Hard disk drives (HDDs) are inexpensive and high-capacity data storage media that do their job in a generally reliable fashion. To support a reliability level based on the five nine concept (99.999%) in telecom applications, however, engineers are evaluating solid-state flash disk as hard disk drive replacements. Solid-state flash disks have no moving parts, so they can withstand the very same conditions that cause HDDs to fail. With the continually declining price of flash disks in ever- higher capacities, they have become a viable option in many telecom applications.

The biggest problem with hard disk drives is their limited reliability due to the fact they have moving parts. Failures come in a variety of guises, such as read/write head failure and motor problems. Overall, HDDs exhibit an average 2 percent failure rate in systems operating in controlled environments. Conditions of high shock, vibration, altitude, humidity and extreme temperature ranges can further boost the hard-drive failure rate percentage to a double-digit figure, unacceptable for mission-critical telecom systems.

Like their mechanical counterparts, solid-state disks based on flash memory retain data when power is off. But unlike mechanical disks, they have no moving parts. This makes flash disks far more reliable, achieving mean time between failure (MTBF) rates 5x as high as HDDs, and a bit error rate (BER) of 10^-20. The solid state nature of flash disks also provides a number of other advantages, summarized in Table 1.

Table 1: HDD vs. Solid-State Flash Disk

Table 1 - Because they do not have moving parts, flash disk drives compare well with hard drives in most specifications, especially reliability.

Meeting System Needs

To serve as a replacement for HDDs, however, flash disks must meet system criteria. One is compliance with NEBS (Network Equipment Building System) standards. NEBS is a common requirement in North America and includes a set of technical specifications (GR-63-CORE, GR-1089-CORE, GR-78-CORE) with one major purpose: to make network switches “bulletproof”. Carriers must prove their equipment complies with NEBS before the equipment can be used in any central office facilities. Flash disks easily meet this requirement. They typically are NEBS Level-3 compliant (carrier class environments with maximum availability) and can be used for data storage in optical switches, ATM switches, IP gateways and core routers, public telephony switches, PXBs and wireless base stations.

To provide true “drop-in replacements” for HDDs, flash disks must also have identical dimensions, the same mounting holes and the same system interfaces. The most common flash disk form factors include 2.5” (laptop size disk) and 3.5” (desktop size disk), with either IDE/ATA or Narrow/Wide SCSI system interfaces. Here, too, flash disk drives easily meet the requirement.

What has been a real barrier in the past to the deployment of flash disks has been their high cost per byte of storage. That barrier is wearing down, however. The flash industry has its own version of Moore’s law: the density of the flash component doubles within the same silicon footprint size every 12 months. This has resulted in a doubling of capacity every year for the same casing size, sharply reducing flash cost per byte.

Even so, flash disks are more expensive to purchase than HDDs. When choosing the storage solution for any application, however, it is important to calculate the total cost of ownership (TCO) instead of merely the initial purchase cost. While HDDs typically have low acquisition costs, factors such as life expectancy (2-4 years), down-time, maintenance costs and warranty period (1-3 years) should also be taken into account. Here, the higher reliability of flash disks reveals its full advantage, resulting in a lower TCO.

Supporting Serial Interfaces

To achieve the data rates of modern telecom systems, storage devices are moving from parallel to serial interfaces. Serial ATA, for instance, is an evolution of the ATA interface from a parallel bus to a serial bus architecture, allowing point-to-point connection and providing faster data transfer rates, hot swap capabilities and easier configuration (Table 2). Many telecom systems are moving their storage interfaces to serial ATA, which will be introduced at 150 MB/sec, with a roadmap planned to 600 MB/sec. This roadmap allows serial ATA to support storage evolution for the next decade, based on historical trends.

Table 2: Comparing Parallel ATA and Serial ATA Interfaces

Table 2 - Many storage systems, including flash disks, are moving to the serial ATA interface to accommodate increasing system data rates.

Flash disk systems support serial ATA. M-Systems’ Fast Flash Disk (FFD™) 2.5” Serial ATA flash disk, for example, provides burst read performance of 150 MB/sec, sustained read/write rates as high as 44 MB/sec, and top data integrity under the harshest environmental conditions. In addition, the disk supports M Systems’ True FFS® flash management, which results in NEBS Level-3 compliancy and a storage endurance of more than 5 million write/erase cycles.

This last factor, endurance, is a characteristic of flash memory that may be unfamiliar to HDD users. All types of flash memory have a specified life expectancy, measured as the number of erase cycles that memory cell in the flash device can undergo. Raw flash memory endurance can range from 100,000 to 3,000,000 cycles, depending on the technology. At first glance, this seems like a severe restriction. Using a flash management tool, however, can substantially increase flash life expectancy to the point where the limit is of no consequence in even the most write-intensive applications.

The key to extending flash lifetime is to avoid using the same memory cells repeatedly. Simple flash file algorithms map a logical sector to a fixed physical location. This method quickly causes the flash to wear out when an application updates the same sectors over and over again, a very common usage scenario. For example, all file systems need to maintain some data that describes the allocation of sectors to files and this data is located in a specified sector of a disk drive. A file allocation table (FAT) file system updates the FAT every time a file is extended or concatenated. The FAT resides in sequential sectors located at the beginning of the media. A simple flash management algorithm thus could suffer catastrophic failures after only several thousands of file operations because it would wear out those memory cells.

Boosting Flash Lifetime

Advanced algorithms for flash management map a logical sector to a physical location but change the mapping over time so that it is not always the same sector. This method, called wear-leveling, ensures that all write and erase cycles are evenly spread across the entire flash array to extend flash life expectancy. Static wear-leveling is applied on static files that are characterized by sectors of data that remain unchanged for very long periods of time. Dynamic wear-leveling is applied on newly written data based on a statistical allocation of units. The combination of both static and dynamic wear-leveling has a profound effect on increasing flash life expectancy.

More than extending flash life expectancy, wear-leveling also delays the onset of certain failure mechanisms in the flash. These failure mechanisms can cause entire erase units to become inoperable. When wear-leveling is used, however, the erase cycle limit of the flash is increased well beyond the minimum specified by flash vendors. Flash vendors such as Toshiba and Samsung commit to 100,000 to 300,000 raw flash write/erase cycles, depending on the operating temperature condition. With its TrueFFS flash management tool, M-Systems is able to guarantee more than 5,000,000 write/erase cycles inside its flash disks. In addition, TrueFFS ensures the highest reliability in a flash disk by mapping out bad blocks, and by implementing both hardware and software-based error detection and correction code (EDC/ECC).

This added reliability is of critical importance in telecom applications. The time, cost and damage in reputation that may result from a faulty disk drive can be devastating. Solid-state flash disks can operate under the harshest environmental conditions to provide top data reliability, enabling engineers to approach and sometimes even achieve the five nine concept (99.999%) reliability standard. And as the density of flash continues to double every 12 months, its cost will continue to decrease, making it more cost-effective as a HDD replacement.

Esther Spanjer is a Technical Marketing Manager at flash memory system manufacturer, M-Systems. She received her B.Sc. degree in Electronic Engineering from the Technical University Amsterdam (Netherlands) in 1991.