|
||
|
|
|
              |
|
 |
 |
||||||||||||||||
|
 |
|
||||||||||||||
By William Van Winkle |
||||||||||||||||
This is not an article devoted to beating you over the head about the importance of backup. If you don't get backing up yet, you have bigger problems than we can address here. Rather, this is an article about storage. With internal SATA-based storage now landing near 25 cents per gigabyte and 1TB drives hitting shelves around the time you read this, the thought of not having some kind of scalable storage architecture seems wasteful if not negligent. Anyone from a school kid through the Fortune 500 needs some form of redundancy applied to every file worth protecting, and who among us hasn't outgrown his or her storage limits—several times? The logic behind storage growth dictates ever-increasing resources, but doing this in a sensible, long-sighted, cost-effective manner needs a reseller's touch. |
||||||||||||||||
Burgeoning Bit Basics
The fundamental RAID concepts emerged in the 1970s, but the term itself didn't originate until a paper was published through UC Berkeley in 1988 called "A Case for Redundant Arrays of Inexpensive Disks". The paper outlined several RAID models, each imagined for different user scenarios, and the list of models, or levels, has since grown considerably. The most common levels today are RAID 0 (striping across multiple drives for faster performance), RAID 1 (mirroring one drive onto another), and RAID 5 (striping data and parity across at least three drives so that if one fails the data can be rebuilt from the remaining units). JBOD, or "Just a Bunch Of Disks," is not a RAID level per se but is often listed alongside RAID; the process refers to grouping multiple disks into a single volume. Those who want performance from their disk subsystems benefit from the faster read/write operations of RAID 0. Those who value data protection veer toward mirroring, since if one side of a RAID 1 fails, the RAID software instantly activates the mirror drive, leaving the user with no data loss and no immediate downtime. In theory, RAID 5 offers the best of both striping and mirroring, but the catch is in the processing power needed to compute and distribute those parity blocks. Additionally, RAID 5 incurs a lot of additional traffic between the drives and controller. Until fairly recently, there was so much compute overhead entailed in RAID 5 that the feature couldn't be implemented into chipset southbridges. RAID 0 and 1 were common core logic check-offs, but RAID 5 would cripple the CPU responsible for crunching that parity data. Software-based RAID cards were little help because the parity was still being handled by the main processor. Only business-class controller cards from the likes of 3ware, Adaptec, and LSI that used dedicated storage processors to offload the RAID burden made reliable data protection possible without smashing system resources. With the latest generation of silicon designed around dual- and quad-core CPUs, though, there's finally enough processing overhead to accommodate RAID 5 in software. Many customers balk at losing half of the drive capacity they're paying for when implementing RAID 1. With RAID 5, though, the capacity equation is simply N-1. Presuming the use of 500GB drives, three drives in a RAID 5 leave you 1,000GB of capacity. Four drives leaves 1.5TB, and so on. Coupled with a fast processor and modern chipset, RAID 5 becomes a much more palatable option, even for home users. Naturally, we don't expect secondary PCs to be RAID 5 havens. Junior high homework machines can have all their protection needs met with a $50 CD-RW or USB flash drive. But once we start talking about dozens of gigabytes of multimedia files, especially the irreplaceable photo and video files that constitute "mission critical data" in the home, scalable capacity and protection become far more relevant in key consumer, SOHO, and even small business primary PCs.
Scalability is key even here at the low-end, because once we start talking about protection, the reseller's job is to equip buyers with an infrastructure that can migrate upward with minimal cost and downtime. Keep in mind that users shouldn't take available capacity as a gauge of their storage needs. According to online utilities site PC Pitstop (www.pcpitstop.com), which keeps anonymous statistics on the drives it analyzes, as of the end of last year, the average home user's hard drive averaged about 145GB in capacity, of which roughly 100GB remained free space. On the business side, hard drives averaged 100GB in total capacity, of which about 75GB was free. So clearly, most users don't seem to be running out of room on their internal drives, but they still have challenges in how their data should be managed. At the desktop level, there are obviously three solutions: add more internal storage; add USB, FireWire, or eSATA direct-attached storage; or add network storage. According to iSupply numbers for 2006, the hard drive industry shipped 15.5% more units than in 2005, but the external hard drive market was up 37% for the same period. The advantage of internal storage is less cost per gigabyte and far greater performance; the downsides are higher installation complexity/cost and that external drives can't be aggregated into RAIDs. The exception is eSATA drives. These offer up to the full 3 Gbps interface performance of internal SATA while preserving the convenience of an external format. Seagate's eSATA External Hard Drives, now available in 320GB and 500GB models, remain the best-known option here, but Iomega, LaCie, and others are joining the field. Moreover, for the budget-constrained, 3.5" SATA-to-eSATA enclosures now cost less than $40. The trick with eSATA is to make sure the motherboard not only supports the format but integrates the eSATA port into the chipset's RAID options. This is a rare feature today, but expect to see a lot more of it as the year progresses. Once eSATA RAID becomes more common and dual-format USB/eSATA drives gain steam, single-format USB or FireWire drives will have a tough time maintaining their value propositions.
For simple, consumer-oriented backup purposes, USB drives are fine. Dual-drive USB externals, such as Maxtor's OneTouch III, Turbo Edition and Western Digital's My Book Pro and Premium, are better since they can be configured to run in RAID 1 mode—no minor point given that many users subject their externals to the rough and tumble of frequent transportation. We favor eSATA for its superior performance and forward-looking compatibility. But network-attached storage (NAS) fills a wholly different need. IDC's Worldwide Disk Storage Systems Quarterly Tracker shows the fourth quarter of 2006 yielding 4.9% revenue growth year-over-year. Much of the segment's health stems from network-attached storage (NAS), which showed 21.9% year-over-year acceleration. Some of NAS's success comes from $150,000 and up enterprise storage systems finally being outsold by mid-level ($15,000 to $149,000) buyers seeing the value in nearline storage in addition to higher-speed transactional storage. But NAS is also getting a push from the bottom end of the market. Solutions here include basic, single-drive units such as SimpleTech's SimpleShare to super-simple, dual-drive enclosures like NETGEAR's SC101T. Also consider value-add drives like Seagate's remote access-friendly Mirra Personal Server or Maxtor's family-and-friends-oriented Fusion for users who not only want the local sharing aspects of NAS but also the ability to extend that sharing beyond the confines of their immediate LAN. And in all cases, putting storage on the network instead of tying it to a particular PC eliminates the risk of hammering that PC's performance when multiple users on the LAN try to access that storage simultaneously.
The beauty of external storage is that while internal drive margins have wasted away to dangerously anorexic levels, external storage still has its pockets of decent profit. This is particularly true with NAS, where you not only get to sell the box but also have the potential to beef up the surrounding network and configure some advanced file access functionality. The trick with consumer storage going forward will not be how to give someone a simple backup target. Backup is only the starting point. Ideally, the storage solution you implement should be future-compatible, deliver more functionality than the user expects, enhance rather than bog down overall system and LAN performance, and scale with the user's needs such that he's not throwing out his last investment every time he needs to expand his capacity.
Survival of the Serial By now, you know all about the asteroid impact of serial technology and the slow, global death of parallel. Parallel printer ports ceded way for Universal Serial Bus. More importantly for this discussion, parallel ATA is now evolving into SATA. And the lumbering giant, SCSI, only survives against Serial Attached SCSI (SAS) through the virtue of its expensive, decades-long install base. An in-depth understanding why serial technology is superior isn't necessary here. Suffice it to say that while serial may have fewer wires, the ability to ditch parallel's old dependence on synchronized clock rates allows serial to blast signals at dramatically faster rates than parallel could ever achieve. That said, today's SATA drives aren't that much faster than their PATA counterparts, and this is because while the burst rate possible across the drive interface is much higher with the latest SATA spec (300 MB/sec vs. PATA's maximum 133 MB/sec), the sustained rates at which data reads from and writes to the drive platter have not substantially changed so far. That shouldn't be taken as an endorsement of PATA. There are too many other advantages with SATA, not least of which are future compatibility and the simplicity of point-to-point cabling.
The command set, programmability, and robustness that made SCSI such a titan throughout the business world carry forward into SAS. Serial Attached SCSI mirrors SATA's present 3 Gbps connection rate; in fact, it's even plug-compatible with SATA. (Whether a controller lives up to the promise of seamless compatibility across both formats is another issue.) Like SCSI, SAS is expensive compared to desktop drives, but in transactional environments, where lightning-fast access times, blazing transfers, and can't-fail dependability are paramount, SAS is the go-to superstar. In the business world, SATA and SAS are coming into increasing conflict thanks to "enterprise SATA" hard drives. Seagate's Barracuda ES and Western Digital's RE series are two standout examples. These deliver the higher reliability of SAS drives but the higher capacities and lower pricing of SATA. SAS controller vendors like to talk about how implementing a SAS controller architecture provides a way to save money with SATA now yet seamlessly upgrade to SAS down the road if budgets and/or system needs change. This is a valid point—and a decent sales tactic—but so far not an approach that many businesses have followed through on. The systems that start out with SATA generally stay SATA. The bigger question is whether or not SATA is good enough for high-performance server needs. In a server context, SATA is for always-available data that has aged enough not to be transactional anymore and is moving to reference. This is what most people mean by "nearline" storage. For example, consider ticketing information before a flight takes off. That's transactional data likely to get sought hundreds or thousands of times in a relatively tight time window. After the flight arrives and all the passengers and luggage have been settled, that data needs to be kept online for reference, but it's not transactional anymore. High-performance access is no longer needed, so the priorities for that data's infrastructure changes. Best dollar-per-gigabyte and long-term reliability take center stage. On those bases, SATA is the clear winner. Can SATA win on performance too? Certainly, there's a difference between 20 and 2,000 users needing data off the same drive simultaneously. But there's no hard line dividing when one needs to move from SATA to SAS. Even if you could pinpoint the exact bandwidth to serve a particular application on a particular network configuration, users and demand will change over time, and admins need to allow for future growth. SAS could seem like the logical conclusion, but you may find that a handful of striped SATA drives provides sufficient throughput with higher capacity. One way to protect this is with a RAID 0+1 configuration, which yields a mirror of a striped disk set. You should also assess Western Digital's Raptor drive as a midpoint between conventional SATA and high-end SAS. Like enterprise SCSI/SAS drives, the Raptor uses a smaller 3.0" platter to achieve its 10,000 RPM spin rate. Additionally, the drive features improved vibration tolerance, a 1.2 million-hour MTBF, a five-year warranty, and time-limited error recovery (TLER). This is an enterprise-class feature that limits error recovery durations so RAIDs don't drop a drive from the array (a preventative feature called "fallout") when it gets stuck trying to repeatedly fetch bad bits. For users who don't need some of the advanced functionality of SAS, a 150GB Raptor's $230 OEM street price compares very favorably against the $400 to $500 you might pay for a 146GB SAS drive with the same rotation rate. "There are certain verticals where we're seeing more gravity towards enterprise SATA," says Western Digital's Darrin Bulick, senior marketing manager for enterprise products. "One of the big ones is video surveillance. Medical imaging is another. In some cases, the Raptor's performance can really pay off. But then there's the nearline market, where you have a repository of storage where things don't need to be served up as fast. This is where our RE and RE2 drives play best." Sometimes the argument between SAS and SATA can't be resolved by a simple look at throughput rates. The issue may come down to abilities such as drive clustering and addressability. Many SATA drives are natively point-to-point devices and have no addressing capability. (Western Digital notes that its SATA drives are an exception.) On the other hand, SAS drives and controllers, just like Fibre Channel devices, use an 8-byte World Wide Name (WWN) identifier that can be critical in having granular control over storage area network (SAN) implementations. That said, there are partial work-arounds to improve SATA's high-end prospects. To add addressing, you have to add electronics to a SATA drive in the form of an interposer, also sometimes called a tailgate. This is a little board with a SATA connector on one side (that plugs into the drive) and a SAS connector on the other. Both the board and drive then plug into a SAS environment. The interposer's electronics give addressability to the drive and also a pseudo-second port that makes the drive clusterable. Interposer vendors such as Emulex (www.emulex.com) and SiliconStor (acquired by LSI Logic last February) may also build in other features, including acceleration of SATA spin-up times and additional error recovery. However, the interposer adds cost while not adding to the drive's reliability, so don't think this is some sort of magic cure-all for desktop SATA drives. Also keep in mind that while the independent ports on a SAS drive both operate at full duplex, the interposer is a single multiplexor, meaning you can only communicate through one port at a time with an outside address. Two hosts can talk to the drive, but not concurrently. Interposers remain a little-known or -deployed solution, most likely because they sit in the gray area where people normally start thinking about moving up into SAS rather than staying with Serial ATA. Companies interested in building a clustering environment to prevent single points of failure tend to be less interested in shaving cost corners. The additional benefits of SAS become more persuasive. ...more |
||||||||||||||||
 Back to top
|
||||||||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||||||
