Last month, we introduced you to some of the most recently released core server technologies. The past year has seen numerable announcements spanning processors, chipsets, motherboards, memory standards, and even bus topologies. Individually, those advancements contribute a little extra performance here and some power savings there, but embraced on a whole they make it possible for you to build better servers with new features that encourage existing customers to upgrade and expanding businesses to invest in hardware that improves productivity.

However, while picking the right motherboard, arming it with a potent processor, and adding sufficient memory might be the key to a great standalone machine, servers have several other critical subsystems that must also be addressed. Storage devices, for example, need to be more reliable and faster than your everyday desktop hard drive since they accommodate the burden of simultaneous requests from multiple clients. Along the same lines, devote some attention to network infrastructure. A great server won’t accomplish much in an environment with shoddy wiring and a low-speed connection. Then there’s the issue of minimizing downtime. Heavily stressed hardware will invariably fail at some point, and it’s your job to implement enough redundancy to gracefully replace fallen components. And think outside of the box. Local backup isn’t effective protection against fire and theft, so formulate a more comprehensive routine for protecting priceless data.

It’s All About Storage

Designing a robust storage subsystem is one of the most daunting aspects of server construction. There are several different types of interfaces, all with varying speeds, and a number of hard drives that correspond to each. But when it comes to server storage for small and medium-sized business, the landscape is, surprisingly, a little more straightforward. You can almost immediately toss out expensive multi-drive SCSI RAID arrays. Intended to cope with heavy duty cycles, 24-hour operation, and massive traffic, SCSI is a bit overboard for the lighter loads that smaller shops generate.

Perhaps the most attractive storage technology right now is Serial ATA. Currently deemed inappropriate for mission-critical enterprise deployment, Serial ATA follows up on the increased penetration of EIDE into the entry-level server market by enabling higher performance and more flexibility in configuration. Unfortunately, Serial ATA 1.0, today’s implementation of the technology, isn’t always the best route for storage. It isn’t designed to coexist in ah chassis full of other vibrating drives, it doesn’t provide the fault manageability of SCSI, and each Serial ATA port only supports one device, limiting expandability.

The second generation of Serial ATA technology, presently emerging through updated hard drives and a handful of chipsets, promises to allay some of those shortcomings. On the issue of performance, Serial ATA II will double throughput per port up to 300MB per second. That might seem like a silly thing to do considering Serial ATA hard drives only push 60MB or 70MB per second, but Serial ATA II will also enable multiple drives to connect to each port through a multiplier. The extra bandwidth is necessary in order to facilitate up to eight-drive arrays on a single channel. NCQ (native command queuing), a feature previously exclusive to SCSI hard drives, improves efficiency during the random accesses typical of server environments. Finally, the same manageability protocols used to diagnose faulty SCSI and Fibre Channel drives are part of the Serial ATA II specification, making it much easier to troubleshoot new hard drives.

Of course, building a server that centers on the latest storage technology requires compatible drives and controllers. Maxtor and Seagate are both expected to ship their second-generation Serial ATA drives later in 2004, and NVIDIA’s nForce4 chipset should emerge in roughly the same time frame with corresponding support built in. NVIDIA takes the technology a step further by implementing its own features that simplify manageability. Software that ships with nForce4 motherboards will allow you to diagnose failing drives without having to interpret an onboard LED. If you want to add fault tolerance, the chipset will recognize a spare hard drive and, in the event of a failure, rebuild an entire RAID 1 array on the fly without any impact on performance.

Neither of Intel’s high-end PCI Express chipsets, E7520 and E7525, works natively with the Serial ATA II feature set, so you’ll want to look into add-in controllers from Marvell, Silicon Image, or Promise when it comes time to configure.
If you’d prefer to de-emphasize using the server as a storage repository, NAS (network attached storage) is a reasonable alternative for customers with greater capacity demands. NAS devices stand alone, acting as dedicated file servers on a network. They center on highly integrated embedded platforms that manage cost and maximize efficiency. Generally, you can expect a NAS server with RAID 0 and 1, its own operating system, 160GB capacity, and a bundled anti-virus solution to cost just over $1,000, which might very well be worth it to an SMB customer with moderate file sharing requirements.

Balancing Performance With An Adequate Network

It doesn’t really matter how many processors you use or how much memory your servers boast if they’re communicating over a dilapidated network. Gigabit Ethernet is integrated onto a vast majority of modern motherboards and many server platforms include two Gigabit controllers for improved security. While most projects won’t explicitly require that much raw throughput (Gigabit is theoretically capable of about 125MBps, though it never gets that high), low cost and ease of implementation make Gigabit Ethernet an easy upsell for longevity and value.

Actual deployment can get a little tricky if your client is in an older building with standard CAT5 patch cables running to and fro. Technically, Gigabit can run over CAT5 without any problem if crosstalk, continuity, and signal loss all check out. But in those situations where it doesn’t, you might suggest looking into CAT5e. Keep in mind that poor splice jobs and lengths that exceed the 100-meter limit of CAT5 will both reduce network performance

There’s a good chance that during your talks about networking, the subject of wireless connectivity will arise. Fair enough, Wireless is a pervasive technology both in the home and in many offices, even if it can’t keep up with the performance of Gigabit. One of the top concerns when it comes to installing and configuring a wireless network is potential security vulnerabilities. WEP (wired equivalent privacy), the original security protocol for 802.11b networks, has already been compromised and the upcoming 802.11i security specification is still being implemented. An interim solution, WPA (Wi-Fi protected access) is much safer, but I’ve dealt with enough gun-shy executives to know that the small business owner is reluctant to assume unnecessary risk.

Should your customer specifically ask for it, be upfront about adding wireless networking to a server. If it’s the utmost in performance they need, go with Gigabit. If there’s any trepidation about security, go with Gigabit. However, I’ve actually found that customers who value flexibility extol wireless. In many cases it’s significantly easier to work through walls instead of around them and the 802.11g standard delivers plenty of performance and a broad upgrade path in the face of a growing business looking to keep workstations connected.

Backup Your Business

After working on individual PCs all day, it’s easy to forget some of the characteristics that make a server stand apart. Next to the components required for a server to function, backup is one of the most important add-ons to consider. Sure, it’s an extra expense, but there are plenty of available devices, allowing your customer to choose the method they prefer.

The most traditional backup mechanism is the tape drive. Centering on mature technology, tape storage costs less per gigabyte then any other medium and the top-end models boast expansive capacities. The tradeoff is that tape drives suffer poor access times, making them less than ideal for pulling individual files from an archive. Moreover, high-capacity drives are expensive, as are automated loaders.

If your customer is insistent on tape storage, DAS (direct attached storage), as opposed to LAN-based, provides the most economical route for management, support, and acquisition. Within the family of DAS devices, there are a plethora of technologies with varying performance and price points. DAT (digital audio tape) drives combine reasonable cost and modest capacity. Unfortunately, they’re only made to work with SCSI controllers and their write speed is fairly disappointing. DLT (digital linear tape) is significantly faster, holds more information, and is generally regarded to be more reliable. It costs more though, and again, is limited to SCSI connectivity. A more flexible option is an AIT (advanced intelligent tape) drive that stores up to 90GB of compressed data, boasts moderate performance, and won’t break the bank. The available USB 2.0 interface makes it easy to use the drive on any number of computers, too.

Optical backup is also a popular choice, though less so for large-scale server archives. A DVD writer costs a fraction of what a tape device runs, and 4.7GB discs usually hover around $1 each. The latest 12x burners write complete discs in about six minutes. Plus, modern backup applications, such as Dantz Retrospect 6.5, are able to span an entire job across multiple DVDs. Saving 160GB of data on 4.7GB is an arduous task, though. The latest double-layer drives help reduce the number of discs used by storing 8.5GB, but at $12 per disc the media is still prohibitively expensive.

If you’re looking for the best balance between performance, capacity, and price, it’s hard to beat an external hard drive. Usually compatible with both USB 2.0 and IEEE 1394, they’re able to transfer at 480Mbps or so and connect to just about any computer on your network. Further, companies such as Maxtor and Seagate sell drives that store up to 300GB. Paired with a good backup app, incremental saves only take a couple of minutes per day.

Redundancy Works

Though the components that comprise a server are built with reliability in mind, there’s always a chance that something could go wrong. Worse, you can’t control the consistency of variables outside your customer’s server, such as stable power. That’s why it helps to minimize the number of influences capable of inflicting downtime. The previously discussed backup solutions deliver one level of redundancy, protecting the server’s most valuable contents.

Redundant power supplies add a second level of defense, maintaining server functionality even if one unit should fail. However, you’ll probably get more value out of a UPS (uninterruptible power supply) battery backup system that can maintain power to a server for a few minutes after an outage. That’s not enough to continue operation, but it should be sufficient to perform a graceful shutdown instead of losing unsaved information in an instantaneous crash. Of course, UPS systems have their own hard-to-decipher specifications, so to make the job a little easier, check out APC’s website (www.apc.com), where you’ll find a configuration tool that recommends hardware based on your server configuration and power delivery needs.

In Retrospect

If the overall performance of servers you sell attracts attention, then the small details invested into configuration will yield actual sales. After all, anyone can build a fast computer. It’s the knowledge of available technologies, helpful suggestions, and support that keeps your customers satisfied. So by all means, pick those processors and be mindful of platform capabilities, but remember that servers also benefit from solid storage subsystems, a regimented backup routine, redundant power delivery, and a network connection capable of keeping up with clients.

Finally, take note of how it feels next time you walk into an automotive showroom. You might know a thing or two about each model and you may have some idea what you’re looking for, but if a salesperson starts talking over your head, there’s a good chance you’ll tuck tail and run for cover. Even if they play their cards right and get you to the bargaining table, extra fees here and there will undoubtedly turn you off. Along the same lines, know the technology you’re selling and use it to explain why your server is the server. Just don’t start talking about memory timings, L1 cache sizes, or the advantages of serial topologies. And come to the table with a package that meets your customer’s needs. Selling a server, tacking on the backup device, and adding a software package after the fact won’t convey the positive message you’re striving to send.