Server Month wrap-upDate: Apr 11, 2011
The five different 1U servers examined in our Server Month coverage and what you should consider when purchasing 1U servers are discussed in this video.
Find out which features stood out for different servers, including thermal management, cooling capabilities and efficiency designs.
The team also discusses the importance of expandability for 1U servers, the time it takes to configure these servers, technical considerations, maintenance contracts and service concerns.
Check out the rest of our Server Month tips and videos.
Read the full transcript from this video below:
Server Month wrap up
Colin Steele: Hi. I'm Senior Site Editor, Colin Steele.
Steve Bigelow: I'm Steve Bigelow, Senior Technology Editor.
Carl Brooks: Carl Brooks, Senior Technology Writer
Colin Steele: We hope you've enjoyed our videos on major one-use servers, and now it's time for some closing remarks. Guys, I was hoping we could start off by talking about some of the characteristics of the different servers that really stood out to you.
Steve Bigelow: Well Colin, it was really the Thermal management that first caught my attention. Most one-use servers have ten, twelve, even more fans that move air from the front of the unit, all the way back through to the rear of the unit. And the flow of air will cross the CPU heat sinks, and in many cases, the memory modules as well. That works together to provide a lot of air flow and goes a long way toward keeping the unit cool. Some designs, especially like the super-micro that actually went out of its way to put the memory modules in line with the airflow and very streamlined in the way that it moved air through the unit. I also was very surprised by the addition of two, ten GB Ethernet ports to the super-micro system. It might make it a serious candidate for Ethernet-based SANS like iSCSI or FCOE applications, especially where there are a number of workloads that need to be accommodated by the unit. I was also surprised by the half-depth SGI server. I thought it was a very interesting way to increase the server density in the rack. How about you Carl?
Carl Brooks: Yes, that's true, cooling was definitely advanced here. One thing is the versatility of it, not only do you have the fans and the airflow and that sort of design, some of these are clearly designed for maximum efficiency, meaning that they move air that cools as much as possible. Some of them are designed for maximum efficacy where more things are cooled and more air might move but it might not be the most efficient cooling. However, you can pick and choose and a lot of these can do multi-duty. Even the monster HP and Dell servers, which are packed full of plastic, they have a lot of options for fan control. You can decide how fast you want them to spin; you can decide how much cooling. It's good; you can set thresholds and alerts so there's quite a bit of versatility here. Even if you're going from something clearly designed to be used in a cluster, like the little SGI or one of the standard one-use servers. That might be the only server in a rack. So you've got a lot of flexibility built on. That actually leads me to the next observation that these things have a lot of connectivity built right on. It's so cheap for server manufacturers to slap a chip on there or put in an add-on card or just build it right onto the board. These things have Ethernet, they have onboard management, they have USB, they have PS2 sometimes --I wish they had more-- they have serial, they've got VGA all over the place. There's just a wealth of options for connecting and interfacing with these things, which is nice to see. It speaks to how commoditized and how cheap it has become to put these things directly onto the board. Not a lot of surprises but definitely some advancements.
Colin Steele: The other big issue that I wanted to ask about is expandability. How important is that with these one-use servers?
Steve Bigelow: Colin, personally I really think expandability is a little bit overrated when it comes to one-use servers. Most organizations, they'll choose the options, the number and type of processors, the amount of memory, the number of hard drives and so on that they want in the server at the time they actually decide to purchase it. So in most cases PCIE expansion slots, they'll be filled with raid controllers or SAN controllers, whatever specific types of expansion devices the server actually needs before it ever gets to the organization or gets into a rack. Other than adding more drives or more memory at some point in the future I don't really think there's much need to worry about expandability, Carl?
Carl Brooks: No, it's something to consider. Some of them are built more open-ended, some of them are definitely built just for exactly what you get them for but in general. A rack server will have the storage it needs on board. Maybe in some cases it's going to have enough that you can use the whole thing as a stand-alone appliance, but mostly you're going to stick whatever your specialized interface card is in the expansion slot. You're going to use the onboard raid and the onboard connectivity and that's about it. That's really what they're for.
Colin Steele: How long should it take to configure these one-use servers?
Steve Bigelow: Well Colin, really for the most part, it shouldn't take more than just a few minutes to configure a one-use server. Like most typical computers, a one-use server these days is very self-contained. You plug it in, you connect to your network cables and you turn it on. The set-up of the system is pretty much preconfigured for a balance of performance and compatibility. So out of the box it really should support just about anything you want to run on it with very, very few exceptions. When you do want to optimize the performance there are a lot of BIOS options that you can turn to, to tweak the performance as long as you're operating system and your applications support it. Carl?
Carl Brooks: Yes, that's right. When you get this thing in the shop you're going to unpack it and put it on the bench and make sure it boots and check it out. Once you stick it in a rack in more or less a couple of minutes, all these things will do PXE right out of the box. They're all going to look out and hunt for a network image to boot from. By default, all of the BIOS options are in more or less either safe or you-don't-care mode. With the caveat that all of these servers have advanced on-chip that is CPU-based virtualization enabling features, which are of some benefit if you architect around some of these features. If you're actually buying a specific server for hardware acceleration go for it but make sure you double-check all of your hardware, your software, your drivers, all of the little bits and pieces or you may find yourself with some surprises you don't want. However, out-of-the-box, even with those features turned on, they're not going to get in the way if you're not specifically using them. By all means check your server before you install it, but yes, they should work right away.
Colin Steele: In this video series we've talked a lot about IO, storage, connectivity, things like that. Are there any other technical considerations that people should take into account?
Steve Bigelow: There is a couple. One-use servers have a lot to offer. They do have a lot of similarities, but really, picking the server, picking the technical details of the server, is only one aspect of selecting the server. What you have to remember about any computer purchase is that it does have to be serviced and maintained periodically. The maintenance contract on the product is going to become a recurring line item every year. As you're making your product selection ,you do also want to think about the service, the support, the warranty terms, how quickly you need to get a technician to the site, so your service level agreement, in terms of support for the systems. Those are often some "gotchas" that go without a whole lot of thought until problems actually occur, and then you wind up with down time. When you're in the purchase mode definitely include some thought to service. Carl?
Carl Brooks: Yes. For sure, as a rule of thumb you want to assume that everything is probably going to work and you want to know exactly what you need to do when something doesn't work. Make sure you double-check and see what you're getting. Some of these servers too clearly designed, I wouldn't call them disposable, but they're definitely designed to be, less thought is going to go into the maintenance than others. You buy one of the HP or the Dells; you might buy support or service along with that. It's a premium product, there are going to be premium options to support it. Buy something that is designed to be one out of one hundred servers in your two or three racks; you're not going to get these same options. At the same time, for those servers and even for the HP and Dell, know what you can keep on hand that are spares, know what you don't need to go back to the company for with your golden handcuffs and know what you can do yourself. The other thing that's notable about these, is that all of them today, buying one of these servers is the equivalent almost of a small data center from five or seven years ago. On one of these one-use servers with two chips you can fit up to 32 cores 192 gigabytes of RAM. That's a couple of racks of servers from seven years ago. So there's a considerable amount of power built right onto things. These things have enough capacity that, for a very modest investment, you will probably have enough computing to run yourself a fairly complicated, fairly intensive small business. By the time you outgrow one of these or two of these you're probably going to be looking at complex architectures like SANS and different sets of server options and things like that so there's a lot of capacity and a lot of power packed right onto this thing. You don't buy ten servers these days. You buy one of these
Colin Steele: We've gone into a lot about technical details and support and service for these servers. To wrap up, do you guys have any high-level strategic advice for our viewers?
Steve Bigelow: Well Colin it's very important to think about the components up front. In most cases it's easier, faster, and less expensive, to determine the needs of the system and buy the system preconfigured with the hardware that's required to meet those needs up front. Rather than having to go back later and upgrade the system. It reduces the downtime. It reduces delays in being able to meet changing workload needs and I think the only real exceptions here are hard drives where you can hot-swap them easily while the system is running. Carl?
Carl Brooks: Yes. For some things this is going to be an easy choice. If you're already stuck in a Dell systems management or iLo, or Openview then the choice is going to be easy. You pick the vendor that has your systems management already onboard, simple. However, for a lot of people that's becoming less necessary. If you use VMware exclusively as a virtualization platform for instance, you might actually be doing less hardware systems management than you used to before. It might be less important, you might also be doing agent base for monitoring or support which could also obviate or help some of these things. There are various types of systems management tools out there that are built to interact with various flavors of servers. What you've got in your shop will determine the relative importance of that. Also, what you're doing with your server, are you fully virtualized? You're going to want something that has a boatload of RAM and has enough extra capacity that you can do the workloads you need. Assume ten virtual machines per server? It's probably a little light these days to assume that. However, if you assume ten virtual machines per server, buy two servers. If all of a sudden you need 40 virtual machines you might find yourself a little cramped so look at where you're planning is. Is your planning around virtual capacity? Is it around actual capacity? Is it around throughput? Is it around storage? What's important to you is partly going to determine what you buy.
Colin Steele: That does it for Tech Targets overview of one-use servers. We hope you enjoyed this series. For Steve Bigelow and Carl Brooks, I'm Colin Steele. Thanks for watching.