Back to Architecture
| As the trends in disk storage technology suggest, the design of a storage infrastructure involves more than the simplistic selection of a topology such as NAS or SAN. If the most fundamental components of storage, disks themselves , are changing and growing more specialized in their feature set, functionality, and price, imagine how much more complexity exists as these technologies are configured into arrays and provisioned to applications! From this perspective, it should be clear that there is no substitute for the intelligence of the designer in selecting and implementing the right technologies to meet the needs of business processes and the applications that support them. Above the level of the disk drive is the issue of disk interface protocols. As mentioned above, protocols are also an area of considerable development. Today, most storage device attachments in the open systems world are made using one of two protocols. The first is the Small Computer Systems Interface (SCSI) and its derivatives such as Fibre Channel and iSCSI. Serial Attached SCSI, Fibre Channel, and iSCSI perpetuate the use of the SCSI command set and other SCSCI conventions but obviate the use of the physical parallel bus interconnect of traditional SCSI. The other dominant interface is Integrated Drive Electronics/Advanced Technology Attachment (IDE/ATA), used predominantly for internal disk attachment in PCs and smaller servers. IDE/ATA is receiving new attention by array manufacturers as Serial ATA (SATA) is brought online as an interface for creating general-purpose arrays of inexpensive disk. Table 6-1 summarizes and compares these evolving standards. Industry insiders and pundits are currently having a field day debating the relative merits of SCSI and SCSI-derivative storage arrays versus ATA and SATA arrays. Champions of SCSI interfaces boast that the tried and true technology provides faster data transmission rates (up to 80 megabytes per second) than IDE/ATA ports and note that SCSI ports enable multiple device attachment, whereas ATA does not. Table 6-1. SCSI and IDE/ATA Evolution
It is worth keeping in mind that, although SCSI is an ANSI standard, it comes in many "flavors" (including a number of variants that are sponsored by vendors rather than codified as formal standards). So, anyone who has managed storage over the past 20 years is probably aware of the fact that two SCSI interfaces may be incompatible, even at the level of connectors. The reach of SCSI has been dramatically extended by wedding the command set to serial transports, such as Fibre Channel. From a 25-m transmission range with traditional SCSI, Fibre Channel extends the distance between devices to 10 km, or to 100 km with special optic transceivers. In addition to distance improvements, serial Fibre Channel operates at much higher speeds than parallel SCSI. Fibre Channel currently offers a 2 Gb interconnect, and 10 Gb links are on the drawing board. Fibre Channel also supports up to 126 devices per arbitrated loop, or a theoretical capacity of more than 16 million nodes in a switched fabric. This capability dwarfs the 16-device-per-channel connectivity capability of traditional SCSI. Substituting a serial channel for a parallel bus is also the thinking behind Serial Attached SCSI (SAS). SAS offers many features not found in traditional storage solutions such as drive addressability up to 4,032 devices per port, and reliable point-to-point serial connections at speeds of up to 3 Gb per second. Through the use of "expanders," as shown in Figure 6-3, topologies can grow to as many as 16,000 devices. Figure 6-3. SAS topology with expanders. Like Fibre Channel and high-end parallel SCSI drives, SAS interfaces support dual porting, a useful feature for fault-tolerant array design. The smaller size of SAS device connectors enable full dual-ported connections on smaller 2.5-in. hard disk drives, a feature only previously found on larger 3.5-in. Fibre Channel disk drives. This is considered very important when catering to applications that require redundant drive spindles in a dense server form factor, such as contemporary blade servers. SAS uses a 64-port expander to take advantage of its enhanced drive addressability and connectivity capabilities. One or more SAS host controllers can connect to a large number of drives, or other host connections or other SAS expanders. Vendors claim that this scalable connection scheme enables SAS to be used to build enterprise-class topologies that can support multinode clustering for automatic fail over or load balancing [4] (see Figure 6-4). Figure 6-4. SAS maximum expander device topology. Finally, iSCSI takes the SCSI command set and operates it as an application across a TCP/IP network. Supporting a Gigabit Ethernet interface at the physical layer, the protocol removes distance restrictions and device connectivity constraints altogether. And by leveraging a true network interconnect, rather than a serial channel interface, iSCSI provides the means to separate storage from servers once and for all and to enable the establishment of storage as its own infrastructure within the client/server hierarchy. In an iSCSI connection, when an operating system receives a storage request from an application, it generates the SCSI command, then encapsulates the command in an iSCSI wrapper packet. This packet is directed to a target device across the network, where it is received, the SCSI commands are extracted, and the SCSI commands and data are sent to the SCSI controller and then to the SCSI storage device. The device response (more SCSI commands and data) are repackaged and transmitted to the requestor using the same protocol. iSCSI was developed by the IETF and became an official standard in February 2003. To date, there are no storage devices explicitly tooled with an iSCSI interface. This necessitates the use of off-drive software and hardware to package and unpackage SCSI commands, which are typically wedded to controllers and host bus adapters. In a sense, iSCSI adds to the existing processing burden of packaging and unpackaging of network message traffic between hosts that accounts for the fact that only about 75 percent efficiency is achieved with IP network messaging (including iSCSI). In other words, a 1 GB pipe dedicated to iSCSI traffic will yield an effective 750 MBps throughput, a 10 GB pipe, 7.5 GBps, and so forth. The balance of the bandwidth is lost to "overhead" processing. In most cases, the overhead burden is irrelevant in actual operation ” assuming that network bandwidth is adequate to application demands. To reduce the TCP/IP stack processing burden on servers, expediting technologies such as TCP Offload Engines (TOE) are being added to adapters, controllers and switch/router blades (see Figure 6-5). At least one host bus adapter vendor is also seeking to offload the iSCSI command packaging and unpackaging stack as well. Figure 6-5. TCP offload engines lessen server CPU load. The bottom line is that high-end SCSI (SCSI 3 and above), Fibre Channel, SAS, and iSCSI interfaces provide the cr me de la cr me of connectivity and array building interfaces for meeting the needs of performance and data intensive applications. Vendors emphasize the dual porting of drives, their manufacture to higher standards of precision and durability, and their flexibility in terms of speed and distance, in arguing for their deployment into demanding enterprise environments. At the other end of the spectrum are inexpensive disk drives, known as ATA drives, originally targeted to smaller systems and PCs. ATA is a disk drive implementation that integrates the controller on the disk drive itself. There are several versions of ATA, all developed by the Small Form Factor (SFF) Committee. Over time, ATA technology has been enhanced to support steadily improving disk drive data rates, but it has consistently utilized a short length parallel bus ”the notorious "ribbon cable" inside every PC ”for connecting one or two drives to the computer motherboard. This reflects ATA's pedigree as an internal storage attachment method. In the late 1990s, vendors perceived that new usage models for PCs and small servers, including digital video creation and editing, digital audio storage and playback, file sharing over high-speed networks, and other data-intensive applications, were placing new demands on hard drive throughput that would require further changes to the ATA bus. To keep pace, manufacturers began work on a serial implementation of the parallel Ultra ATA interface that promised to extend the roadmap for ATA beyond the theoretical limits of the Ultra ATA bus. Serial ATA builds upon (or replaces , depending on who you speak to) the latest revision of the Ultra ATA specification in development by the ANSI- backed INCITS T13 committee, the governing body for ATA specifications: ATA/ATAPI-7 specification, an update of the parallel bus architecture that provides up to 133 MBps (see Figure 6-6). Serial ATA (SATA) introduces a four-wire cable replacement for the 40-wire ribbon cable of parallel ATA and delivers 150 MBps throughput with a roadmap of future enhancement out to 600 MBps over the next 10 years. [5] Figure 6-6. Parallel and serial ATA. Where Ultra ATA technology supported up to two drives ”a master and a slave ”per channel via a shared bus, serial ATA uses a point-to-point connection topology, meaning that each source is connected to one destination. Each channel has the capability to work independently so that there is no contention between drives and thus no sharing of interface bandwidth. This connection strategy also negates the need for master/slave jumper settings on devices. Other technical advantages of the SATA 1.0 interface cited by vendors [6] include:
The capabilities in SATA 1.0 set the stage for a second-generation standard development process, SATA II. Announced at the spring 2002 Intel Developers Forum (IDF), a working group comprised of APT, Dell, Intel, Maxtor, and Seagate began developing a Serial ATA II specification, described as a superset of Serial ATA 1.0. This specification is being developed in two phases to meet market demands. Phase 1 improves the use of Serial ATA devices in server and network storage applications. Phase 2 adds additional features for the entry-level and mid-range server segment and provides the second-generation speed increase (from 150 MBps supported by Serial ATA 1.0 to 300 MBps) for both server and desktop. This effort parallels the development of Serial Attached SCSI (SAS), which has, as a stated development goal, compatibility not only with SAS drives and devices, but also with lower cost-per-gigabyte SATA drives. In late January 2003, the SCSI Trade Association (STA) and the Serial ATA (SATA) II Working Group announced a partnership to enable SAS system-level compatibility with SATA hard disk drives. This collaboration, as well as cooperation among storage vendors and standards committees , promises to facilitate the definition of compatibility guidelines to aid system builders, IT professionals, and end-users to better tune their systems to optimize application performance and reliability and to reduce total cost of ownership. The anticipated result of this parallel effort is that products will begin to appear in the market by 2004 that support both drive types in the same array (see Figure 6-7) or topology (see Figure 6-8). [7] Advocates argue that this will provide flexibility to system builders to integrate either SAS or SATA devices, capitalizing on reduced costs where it makes sense, and consolidating two separate interfaces into one. Figure 6-7. Building a fault-tolerant RAID array with SATA and SAS. Figure 6-8. SATA, SAS, and Fibre Channel: One View of the future enterprise storage landscape. Vendors of early products in this space are keen to assert that most storage requirements can be met adequately by SATA (and in the future, SAS/SATA) arrays. They claim that companies are paying dearly for high-performance parallel SCSI or serial Fibre Channel platforms when cheaper alternatives are now coming to market. |
EAN: 2147483647
Pages: 96