|
||
|
|
|
              |
|
 |
 |
||||||||||||
|
 |
|
||||||||||
By William Van Winkle |
||||||||||||
“If I’m remote,” says Leavitt, “like out at the golf course, and I get a ping that the server’s down, that used to mean my weekend was gone. Now, I just pull out my iPhone, sit down on the course, get into the Web-based GUI, take the virtual drives, move them to a different server in the box, reboot, and that’s it. I’m five minutes behind on my golf game.” Additionally, once you’ve made virtual drives, Intel’s software lets you map any or all of them to given compute node slots in the chassis. This way, in the event of a downed node, all the user or servicing reseller has to do is swap the node, and the Modular Server will take the new node and automatically connect it to all of the virtual drives that were pointed to that chassis slot. This is part of why having a “diskless server” is such an important innovation. Compared to the usual morass of reimaging and/or reconfiguring drives inextricably linked to a blade’s integrated storage controller, Intel’s SAN-based approach is a quantum leap in ease of use and uptime assurance. No longer does a bed of drives tied to a second blade have to be sitting around idle as a redundant fail-over spare.
This same idea applies to server migration. When you migrate a server application, the process is normally sensitive and time-intensive, requiring many hours for imaging and/or reinstallation in the face of a motherboard/CPU upgrade. With the Modular Server and its SAN implementation, when it comes time to upgrade servers, it’s simply a matter of pulling out the old node and plugging in a new one. There’s your server migration. The rest is automatic slot/storage pairing that happens behind the scenes. At launch time, there are only two SKUs associated with the Modular Server product. (CPUs, memory, and hard drives are all separate.) The first is the MFS5000SI compute module (also available in a bulk pack of three), equipped with two Xeon sockets under low-profile passive heatsinks and eight FB-DIMM slots. The second SKU is the MFSYS25. This is the Modular Server chassis configured with a 14 x 2.5” drive cage. The unit includes five server node blanks, the main and I/O fans, two 1000W power supplies, one enclosure management module, one storage management module, and one Gigabit Ethernet switch module. Options include a second storage controller, a second GbE Ethernet network switch module, Gigabit Ethernet I/O mezzanine card(s), and additional/spare power supplies. “The max chassis power here is 3,000 watts,” says Intel’s Young. “That’s well over a single circuit’s 1800W, which is still more than what two 1000W power supplies could potentially pull. If you want to do it with two circuits, you need 20A 110V circuits. Now, the problem is that when you have two power supplies on one circuit and you lose the circuit, not only are you going to lose your redundancy but also two power supplies, which means some of your system is going to shut down if it has to. So it’s recommended that you use four independent circuits. But potentially, if you’re pulling a 220V line, you can run it all off of one 20A circuit.”
Like Supermicro, Intel figures power supply utilization on an N+1 redundancy scenario. With only one compute node, you need one power supply plus a second one for redundancy. With three servers, the Modular Server needs two power supplies plus one redundant. For the whole enclosure fully loaded, you need three power supplies plus the fourth spare. The only other option worth mentioning is Intel’s mezzanine expansion card. Each compute node features two Gigabit Ethernet ports, both of which connect to the primary Ethernet switch. That’s 2 Gb/s output from each server. There are 12 Gigabit Ethernet internal links-—two to each node-—and 10 full-duplex GbE uplink ports on the switch. In cases where the buyer wants to add the second Ethernet switch for either additional Ethernet, I/O, or redundancy, you would need to install the Ethernet mezzanine card for the second switch to connect to. All in all, this is the most complicated thing associated with Modular Server configuration. With very few exceptions, the entire system is toolless and just snaps together.
The thing we like most about the Modular Server is that it obviously was designed for SMBs. There is no equivalent SAS-based design anywhere in the enterprise world. Even Supermicro’s OfficeBlade differs from the higher-end SuperBlade primarily in its number of supported CPUs. The basic architecture remains the same from enterprise to SMB markets. Intel’s approach was thought from the ground up with SMBs in mind but couldn’t move forward until SAS expanders became a market reality. The end results seem worth the wait, though, and the cost benefits to buyers are substantial. “The Modular Server does not require additional software and additional management or infrastructure to integrate it easily,” says Bryan Young. “I mean, everybody that has started off with blades had to go through a lot of heartache. Then once you’re in, you’re in. There’s no getting out. You can’t go back. That’s kind of the point of blade architecture, to lock you in. I don’t believe you’re locked in with our model. You’re still open and flexible, and you don’t have so much money invested in buying other things that you can’t take advantage of the expandable architecture. We maximize the investment so buyers can afford to complete their solution.” “Our cross-over cost is three 1U rack servers,” adds Intel’s Leavitt. “So if you bought three rack servers with RAID cards and redundant power supplies, it’s the same price as the Modular Server with three server nodes. When you get to the fourth server, we’re cheaper. But throw a storage box on top of that—-now we’re even less than three servers! Throw in remote management, remote KVM, the GUI, all these things. The value just keeps adding up and up.” BRING ON THE BLADES
The break-even point on blades just a few years ago was two filled 42U racks. With that cross-over point now dropped to potentially less than three servers and small businesses planting a server for every 10 or fewer employees, who can argue that the time for blades in the SMB market hasn’t arrived? If you’ve got a data center needing 300 servers, absolutely—-go with a more traditional blade architecture designed around fast, very vertical scaling. One of the Modular Server’s few drawbacks is that you can’t daisy chain enclosures (so far). A higher-end design from HP or IBM, especially one decked to the nines in virtualization technology from VMware or the like, is going to fit much better here. However, for a small business or branch office that needs, say, four servers, or maybe even a dozen or so, one admin can manage two or three OfficeBlades or Modular Servers quite easily. Because of this, resellers now have another option for crafting a high-margin, low-expense recurring revenue service. Monthly or annual administration of these boxes is a no-brainer, plus you’ve got the widely believed stigma that blades are hopelessly complex enterprise machines beyond the understanding of small businesses. Supermicro’s remote admin tools are nearly as friendly as Intel’s (albeit without the slick SAN aspects). If you can tackle managing vPro desktops, you can certainly manage blades.
In some situations, the remote management aspect of these SMB blades may be enough to persuade buyers, even those without more than 8U or 10U to control. Traditional rack setups require local control via a multi-port KVM switch built into the rack. Switches with slide-out LCD displays, such as Belkin’s 8-port Titan 17, run over $2,000, plus the admin has to get up, walk over, and be in front of the rack, thereby ignoring whatever else might be going on at his desk. The SMB blades we’ve described eliminate both this extra expense and potential productivity drain. “The other market you could go after here is personal clusters,” notes Intel’s Leavitt. “If you go talk to a professor doing some HPC application, this is just one box. They plug it in, load their application, and go. There’s no wiring or anything. It’s a personal cluster solution. We’ll actually have an enclosure you can put the system in that makes it quiet enough to put right next to your office desk. We’ll call it the Acoustic Pedestal Kit at the end of Q1. The system’s already 65 dB when fully loaded. This quiet kit will take it down to 57 dB and make it better for offices.” Our guess is that the potential market and applications for SMB blades is still largely unimagined. Noise envelopes will continue to fall in well-planned designs as compute power and drive platter capacities rise. The era of 1U and perhaps even 2U machines may be coming to an end, and resellers who see the role that SMB blades can play in this transition stand to profit considerably. Take these two pioneering designs as your starting point. Slice in and carve your niche. |
||||||||||||
 Back to top
|
||||||||||||
Copyright © 2008 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||
