![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
|
|
![]() |
|
|||
By Chris Angelini |
||||
Whenever I engage a new customer in a plan to upgrade or replace hardware, I generally offer a couple of different options. Not that it usually matters—you can cast one look around an office loaded with HP machines and take a good guess that you'll be reselling HP servers. Nevertheless, I like choice and happen to think my organization's machines have some notable advantages over the tier-one boxes I've deployed. So I make sure that the topic comes up in the planning phase. Sometimes I get a receptive audience willing to talk about the virtues of a white box, sometimes the customer confirms my original guess that they're happy with another vendor's hardware infrastructure, and sometimes I get this look like I just tried to sell a used Taurus at a Lexus dealership. In the first scenario, I'm happy as can be to explain my product and the benefits it offers. In the second, I can certainly understand when a customer is comfortable with one vendor's hardware and isn't looking to sway. But it's that last group that really bothers me—the SMBs who think, for one reason or another, that a white box server or workstation is somehow inferior just because it doesn't sport a well-known sticker. Don't those folks know that the technology under the surface is the same? Could they possibly have had a negative experience with a reseller's white box in the past? It's Not as Easy… In asking around, I've found that one common misconception is that any 16-year-old with a CDW account can build white boxes. And to be completely honest, there are admittedly more unprofessionally-built white box desktops out there than name brand systems. But as you start shifting from mainstream machines to more complicated servers and workstations, the barrier to entry starts getting more difficult to overcome. Most prominent, in my opinion, is the sweat equity resellers put in to learning about server and workstation technologies. Value is added not by lower prices or bundled software or even some equation that suggests performance per watt of power consumed. Every vendor has access to the same hardware, after all. Rather, you add value by selling a balanced build—and it's not as easy as it looks to build a high-end system with equilibrium between processing horsepower, I/O throughput, memory performance, and graphics alacrity. I've seen plenty of tier-one workstations running with a single channel of memory on a dual-channel board or 1.5 Gbps signaling rates from hardware rated at 3 Gbps. Customers bring in their two-month old desktops wanting a graphics upgrade because a favorite game isn't running fast enough, only to discover that the vendor used a motherboard without AGP or PCI Express x16 connectivity to save a couple of cents. So much for value. Building the best servers and workstations is a truly thought-intensive exercise. This month, I was reminded how hard it can be by a friend of mine who, like many, started with a goal and procured hardware around the application. Unfortunately, unforeseen "gotchas" started surfacing right off the bat, pushing a build that should have only taken a few days back more than a month. He lost both time and money as hardware had to be swapped out and rebuilt. Frustrated as he was, a few lessons were definitely learned by the time everything was put together correctly. Mainly, despite what your customers might think, it takes lots of homework to be a competent system builder. And even then, there's no substitute for tried and true experience. Incompatibilities happen, and knowing which components play nice together is often a matter of trial and error. I recently had the chance to sit down with him and talk about the specifics of what went wrong. Hopefully, some of his more experience-oriented lessons can be avoided by resellers willing to put in the time learning about the newest server and workstation technologies. Lesson #1: Not All Platforms are Created Equally You buy a regular old ATX desktop motherboard. You buy a standard ATX case. You order up a vanilla ATX power supply. Time and again, they all work together flawlessly. That's the whole point of standardization. And while higher-end platforms are standardized, as well (you'll often find Extended ATX motherboards and cases to match), my friend's first problem emerged when he tried to drop a Tyan motherboard into an Intel server chassis. Size-wise, everything was fine. Screw holes matched up and the power supply came built-in, so no problems there. However, the set of leads you'd normally connect for the power button, reset button, hard drive LED, and so on were molded into a single block coming from the case. Naturally, none of the pins on the motherboard matched up. To be specific, this wasn't the fault of either Intel or Tyan. Intel designs its cases to match its motherboards, which are quite often sold in convenient, platform-oriented bundles. Tyan does the same thing and has its own system for attaching leads. Thrown together, the two just don't get along that well. So right away, this particular builder was looking at either hacking the block into smaller pieces or moving on to another motherboard. Chopping up the individual connectors sounded like a fairly primitive solution to what should have been a more elegant installation, so the motherboard was returned in favor of something from Intel. After all, an Intel motherboard and Intel chassis should go together like peanut butter and jelly. They did, of course, but another issue quickly arose.
Lesson #2: Server and Workstation Chipsets are not Interchangeable I always thought it was interesting that chipset vendors would simultaneously announce server and workstation chipsets, yet most of the specifications matched exactly. The same front-side bus speeds, the same memory technologies, and identical I/O controller interfaces grace all of Intel's newest 5000-series chipsets, just to illustrate. Of course, when you get your hands on the motherboards built on those chipsets, the distinct purpose of each becomes clearer. Server boards are most often characterized by multiple processor sockets, support for lots of memory, and, most notably, the expansion slots that accommodate super-fast network controllers, storage interfaces, and low-latency interfaces such as InfiniBand. Commonly missing are slots for graphics cards. You'll instead find a low-end, stable video core glued right onto the motherboard. Clearly, any machine tasked with graphics processing needs something else. Workstations, on the other hand, tend to look a lot more like turbo-charged desktop PCs. Sometimes they have multiple processors, and plenty of memory is certainly a must. However, when it comes to connectivity, that graphics slot is an absolute priority. Workstations edit video. They render 3D images. For those tasks, PCI Express x16 is the only way to fly. You've probably already guessed what went wrong. The second motherboard, meant to replace the first, housed Intel's Xeon processor, supported DDR2 memory, and featured plenty of connectivity, but it lacked PCI Express x16. Since the system was to be a workstation, back the board went. No way would a server chipset do the trick. Lesson #3: Even When You Think you Know, You Might Not My friend and I sat down with the goal of finding the right board for his application. We now knew it needed to be workstation-oriented, so the E7525 (Tumwater) or 5000X (Greencreek) platforms were probably preferable. He already had the Socket 604 Xeon chips, but by this point Bensley had launched, so we figured an upgrade would be well worth the effort. Greencreek it would be. I popped Tyan's page open while he perused Supermicro. Both vendors featured attractive boards that featured the PCI Express slots for which we were looking. Something wasn't right, though. Whereas Supermicro's X7DA8 advertised two PCI Express x16 slots, one operating at x16 and the other at x4, Tyan's comparable Tempest i5000XL touted a pair of x16 slots with the pair actually running at x8 signaling rates. As a reseller, which should you use? In one case, you're selling the ability to connect a pair of graphics cards, though the second is somewhat handicapped. In the other, both slots take a slight performance ding. If you so happen to be selling an expensive workstation card, as well (my friend is using an ultra high-end Quadro FX 4500), and your customer sees the x8 mode in his NVIDIA driver panel, you'll probably be getting a phone call wondering why that $2,000 card isn't running as fast as it should. Lesson #4: Technology Changes…A Lot Almost every piece of this workstation had been swapped by the time we reconfigured. The original processors obviously wouldn't work. The Kingston DDR2 modules purchased for the other box didn't fit, and we once again faced a motherboard swap. Out went the Intel server board. In went Tyan's workstation replacement. Out went the Kingston DDR2 modules that were purchased, and in went some Kingston FB-DIMM replacements. Out went Intel Irwindale-based Xeon chips and in went a couple of sweet little Woodcrest processors. The basic lesson here was that the constant changing of technology can always throw you a curve even if the latest and greatest enables more value. I've already been faced with several questions on whether to go SAS or stick to SCSI, the market's leading storage technology. Customers voice trepidation about PCI Express when they still see so much PCI-X hardware out there. Swift shifts to newer technologies can help give the reseller a leg up on larger builders. SAS is the perfect example. SCSI persists because of its install base. But when an SMB is starting from scratch, a SAS controller and mixed environment of SAS and SATA drives works great to keep costs down while ensuring scalability. VARs should use the constant flux as a means of outpacing everyone else. Lesson #5: Improve Where Others Fall Short As the system build came to a close, I sent my friend along with some suggestions on how to step things up a notch and address the concerns I've fielded from SMBs debating the white box route for servers or workstations. Number one—and perhaps the trickiest to address—is the issue of service. What happens when a hard drive goes out on a workstation Friday night at 7:00 PM? If it were a desktop, the situation probably wouldn't be critical. But a machine expected to be up all the time needs to be fixed right away. I'm actually facing that very situation currently. By complete fluke, a customer's brand new RAID 1 array went critical due to a dead drive over the weekend. Had I not immediately helped him out, his next server purchase probably would be from a tier-one with available same-day service. The reality is painful, but white box machines are really best paired to top-tier service. Next up is the issue of longevity, which I try to address through intuitive planning. Most SMBs aspire to grow, of course. Thus it's the reseller's job to build a machine capable of withstanding more data and the demands of new software. Whereas many tier-one vendors will start customers out with 80GB hard drives and 512MB of memory in an attempt to keep costs down, I've worked with enough "default configurations" to know that they're only good for a year or two before it's time to throw more money at them. If you're building with the next five years in mind, there'll be an upgrade to the next version of Windows Server to consider at the very least. Building smart helps, as well. Modern server platforms utilize at least two channels of memory, and buying high-capacity modules adds a premium. On a board with eight memory slots, populate four of them with 512MB modules instead of buying a single 2GB stick. The customer gets a better price while realizing maximum bandwidth. And here's another hint: The Bensley platform is the first to support quad-channel memory. So exposing the full benefit of dual, independent front-side buses and dual-core processors actually requires four modules at a time. No Substitute for Experience A mix of research-driven book smarts and real-world experience goes into every server or workstation I build. Without the former, it's impossible to sell the virtues of dual-core processing, FB-DIMM memory technology, SAS storage, and so on. Without the latter, there's no way to know which motherboards work best with your favorite chassis. With a little more wrench time, I'm confident my friend's next project will go more smoothly. And just think—there are some SMBs out there who think just anyone can throw together a white box. If only they knew. |
||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||