Markets, S. Zaffos Research Note 31 March 2003 ATA Disks Redefine RAID Price/Performance Cost-optimized storage infrastructures should include redundant arrays of independent disks built with low-cost Advanced Technology Attachment disks. Core Topic Hardware Platforms: Storage Systems Key Issue How will storage systems evolve during the next five years? Tight user budgets, competition and margin pressures have led a number of high-end and startup storage system vendors to introduce arrays that use inexpensive Advanced Technology Attachment (ATA) disks for back-end storage. These vendors have made system build cost a design objective rather than an artifact of the design. They have also redefined the performance and bulk capacity segments of the storage market, by delivering inexpensive redundant array of independent disks (RAID) systems that are interoperable with their high-performance flagship systems and managed by the same tools. Lowering Costs ATA RAID systems build off a vendor's prior software investments, which can often account for 80 percent, or more, of the research and development cost of developing a new storage series. This reuse of code improves time to market, guarantees interoperability, and lowers the cost and risks of developing a new system. As system configurations grow, disks account for a greater percentage of the total system cost. Hence, moving to lower-cost ATA disks can dramatically lower the costs of large RAID systems. Performance/Throughput ATA-based systems cannot match the performance of likeconfigured Fibre Channel (FC)- and Small Computer System Interface (SCSI)-based systems supporting workloads that are "cache-unfriendly" or have a high write content. Differences in throughput between ATA and FC/SCSI disks are primarily the result of ATA disks' slower rotational speeds and slower seek times. ATA disks generally rotate at 5,400 or 7,200 revolutions per minute (RPM), whereas FC/SCSI disks rotate at 10,000 or 15,000 RPM. ATA average seek times can be twice as slow as those on an FC/SCSI disk. Slower disks provide lower maximum Gartner Reproduction of this publication in any form without prior written permission is forbidden. The information contained herein has been obtained from sources believed to be reliable. Gartner disclaims all warranties as to the accuracy, completeness or adequacy of such information. Gartner shall have no liability for errors, omissions or inadequacies in the information contained herein or for interpretations thereof. The reader assumes sole responsibility for the selection of these materials to achieve its intended results. The opinions expressed herein are subject to change without notice.
input/output (I/O) rates and slower response times for cache misses and writes. Differences in data transfer times are not usually a major component of a system's I/O response time. Queue times and mechanical delays are usually major factors, with rotational delays often being a dominant element. Also, ATA disks can only handle one operation at a time, while FC/SCSI disks can queue and optimize dozens of simultaneous requests an important factor in online transaction processing (OLTP) database processing. Where workloads are "cache-friendly," cache can be a great performance leveler. For example, take two like-configured systems. One system is configured with ATA disks; the other with FC/SCSI disks. I/Os serviced out of cache have a response time of 5 milliseconds. Cache misses, I/Os that are serviced by the back-end disk, have a 20-millisecond response time on the FC/SCSI disk system and a 30-millisecond response time on the ATA disk system. If the workload has a 50 percent cache hit ratio, two I/Os would require 25 milliseconds on the FC/SCSI system and 35 on the ATA system, or 40 percent slower. However, if the workload was cache-friendly and had a cache hit ratio of 80 percent, the ATA system would only be 25 percent slower. Data Availability Systems using ATA disks should deliver levels of data availability that are comparable to those of FC- or SCSI-based RAID systems. This is because RAID makes differences in disk failures noncritical events. A disk fails, the system notifies the vendor, data is rebuilt onto a spare disk or field-replaceable unit (FRU) and the failed disk or FRU is replaced. All of these events are transparent to the server and hosted applications. However, until the market validates that these systems do not have any single points of failure, that microcode updates are nondisruptive, that maintenance and upgrades are nondisruptive, and that data rebuild times and performance and throughput during data rebuilds are acceptable, users are advised to demand guarantees as appropriate. Scalability, management tools and service/support also are not differentiators between a vendor's FC/SCSI- and ATA-based systems. This leaves performance, throughput and price as the main areas of product differentiation within a vendor's offerings. Market Positioning The lower throughput and performance of ATA-based systems suggest that vendors should position these systems into less demanding environments, such as long-term content storage or as a first- or second-level backup storage. As backup devices, 31 March 2003 2
they promise more-reliable, faster backups and lightning-fast recovery times. Backups are cache-friendly applications that occur at native disk speed. In addition, because the production volumes and backups are RAID-protected disk storage systems, configuring the backup software to match the speed of the production and backup copies is easier. However, as users develop a better understanding of ATA-based system throughput and performance, it is likely that these systems will be deployed in some online environments. This is because the low cost of ATA disks makes it cost-effective to configure ATA-based systems that are not like-for-like with their FC/SCSI counterparts. Users of ATA disk-based systems may choose to configure a RAID-1 or RAID-10 back end instead of a RAID-5 back end to deliver superior performance at a lower price point. Enterprises can also choose to install lower-capacity ATA disks instead of higher-capacity FC/SCSI disks to increase the number of physical disks supporting cache-unfriendly workloads or workloads with high bandwidth requirements. Vendor Approaches Four large storage vendors EMC, Storage Technology (StorageTek), Network Appliance (NetApp) and Quantum that are currently active in this market have each taken different approaches to delivering ATA RAID systems. These differences reflect a desire to leverage off their installed bases, protect current product revenue, and extend their brands within their end-user buying centers. EMC: EMC has positioned its Centera as content-addressable storage (CAS), a new class of storage designed to store electronic business documents, images like X-rays and magnetic resonance images (MRIs), video, and so on. CAS offloads storage management from the server by enabling applications to access a file using a unique identifier calculated by Centera using file attributes. This greatly simplifies storage administration, but at the cost of giving up the ability to tune file placement. A Centera cluster can be configured with up to 10 Ethernet connections and up to 307TB of raw capacity. A Centera domain can include as many as seven clusters. EMC claims that Centera has no single points of hardware failure and that clusters in a domain can be located at different sites to provide protection against site disasters. However, prudence would dictate demanding availability and performance guarantees until these capabilities are market-validated. On 12 March 2003, EMC announced CX support for ATA disks. The primary differences between CX systems configured with 31 March 2003 3
ATA and FC disks are in the areas of performance, throughput and price. Initial indications are that a CX system configured with ATA disks delivers about half the throughput of a CX configured with FC disks when supporting highly sequential workloads: workloads representative of backups. CX systems configured with ATA disks use the same management tools and offer the same functionality as CX systems configured with FC disks, including SnapMirror and MirrorView. StorageTek: The StorageTek BladeStore is a direct-attached storage or storage-area-network-attached RAID storage system that is compatible with the StorageTek/LSI Logic D series. This compatibility translates into immediate BladeStore support by Veritas Software's NetBackup and Legato Systems' NetWorker. BladeStore highlights include: 1GB mirrored cache on the controller, 96MB of cache on the front of each blade, eight 2- gigabit FC paths to the host, RAID-0,1,3,5, hot-swap FRUs, and global spares. BladeStore hardware will scale from 4TB to 160TB, although, in the near term, the maximum blade configuration will be 8TB arranged as 10 blades, each with five drives in a 6U (10.5-inch) rack behind a single controller. Expect value-added features like Santricity Snapshot and Remote Volume Manager to become available as StorageTek/LSI enhance the BladeStore offering. NetApp: The NetApp NearStore family of file servers is functionally compatible with other NetApp Filer products. A detailed functional comparison of the NearStore and Filer appliances reveals that NetApp has withheld certain Filer functionality, such as active-active clustering and SnapManager for Exchange, and partnered with backup, archiving and hierarchical storage management vendors to position NearStore appliances in their intended roles. NearStore appliances are offered in configurations starting at 7TB and scaling up to 96TB. Quantum: The Quantum DX30 is a RAID storage system that presents itself as an automated tape library (ATL) P1000 with two to six Digital Linear Tapes (DLTs). This eliminates the need for data centers to modify backup/restore procedures to accommodate the DX30. Alacritus Software, Intransa and Nexsan Technologies are three startups leveraging the cost advantages of ATA disks into the storage marketplace. Alacritus' Securitus I is software that converts an inexpensive server and disk storage into a virtual tape library appliance. Intransa is building an ATA RAID storage system that supports the SCSI Over IP (iscsi) protocol. Nexsan is offering two products: its ATAboy RAID storage systems and its ATABeast, intended for use as nearline storage and as a tape alternative. 31 March 2003 4
Acronym Key ATA ATL CAS DLT FC FRU I/O iscsi MRI OLTP RAID RPM SCSI Advanced Technology Attachment Automated tape library Content-addressable storage Digital Linear Tape Fibre Channel Field-replaceable unit Input/output SCSI Over IP Magnetic resonance image Online transaction processing Redundant array of independent disks Revolutions per minute Small Computer Systems Interface Bottom Line: The low cost of ATA RAID storage systems will encourage their use for long-term content storage and as reliable backup storage delivering fast restore times. Enterprises that understand their throughput requirements and are adept at negotiating with their vendors will have them bid ATA, modular and monolithic solutions to ensure the best possible pricing. 31 March 2003 5