RAM | Reseller Advocate Magazine: Issue 75

	Page 1
	Page 2
	Page 3

By William Van Winkle

[ BLADE BEGINNINGS ]

The X38ML is also a popular pick with third-party vendors for blade solutions. In a sense, blades are the only server you can stick in a rack that occupies less than 1U of space. Supermicro’s SuperBlade enclosure, for example, fits up to 10 blades in a 7U form factor, and that’s before factoring in other enclosure elements, such as LAN switches or passthroughs, redundant power supplies, remote management controllers, and InfiniBand switches. The Intel blade for this system can fit up to two quad-core Xeons and four 2.5” hard drives; the AMD blade can fit up to four quad-core Opterons but only one 2.5” drive. That’s up to 160 cores in 7U of rack space—impressive by any count—but there’s obviously a narrow application band that can make proper use of such high CPU densities.

This high-density model has been tied to the blade form factor for a long time, and historically it would take a full rack or two of blades to reach price parity or better with the same number of servers in a 1U form factor. Recently, though, we’ve been seeing innovation in the SMB blade space, something that would have been an oxymoron only a year or two ago. Supermicro is slanting its SuperBlade system into a quieter, more energy efficient package called the OfficeBlade.

Bring On The Blades
Blade servers are gaining traction in the SMB world. Supermicro was among the first channel-friendly players to arrive with slim blades like these, able to pack two Xeons or four Opterons into one compact module.

Intel just unleashed a wholly new SMB product called the Intel Modular Server With Multi-Flex Technology. The 6U Modular Server is notable for many things, but chief among them is how it decouples storage from the compute resources. The enclosure features 14 2.5” drive bays tied into a SAS expander, which is in turn fueled by the LSIbased RAID storage controller module. In fact, the Modular Server integrates a SAN that can be easily expanded via the storage module’s external SAS connector. The six compute nodes can’t be called blades because they have no directly attached storage. Instead, they boot from volumes within the SAN. This is important because if a compute node goes down, the admin simply swaps in a new node. The enclosure management module knows that the node bay maps to a given boot volume on the SAN, so there’s no additional reconfiguration needed, and the RAID protected storage resources were never in danger from the blade’s failure.

[ THE OTHER "U"S ]

Overall, between pedestals with increasing core counts and blades with dropping system costs, the long-term viability of the 1U form factor seems shaky. The 2U design looks to be a different animal. Between 2U and 5U, the motherboards are very similar. You tend to get one or two riser slots in 1Us, but there’s obviously no vertical room for add-in cards. With 2U, there’s just enough clearance for a half-height card, and that opens up a wealth of options for networking and storage controllers. Additionally, 2U systems are more likely to have a greater number of hot-swap drive bays. Six bays seems to be the most common number, although Tyan’s 2P Transport TA26 hits eight, and Intel’s underappreciated SSR212MC2 finds room for 12—plus another two 2.5” internal boot drive bays!

Bigger Can Be Better
1U servers may be sexy, but 3U and 4U designs can deliver a lot more scalabilty. This Intel S7000FC4UR is a 4U, quad-processor platform with loads of room for full-height card expansion.

“From 1U to 5U, the form factor decision is largely based on the application,” says Intel’s Brian Jarvis. “For example, take hot swap drive support. If that’s important and you need more than three drives, good luck squeezing that into a 1U solution. We happen to sell a lot of 2U servers for that reason.”

Moving up the ladder, 3U chassis add more of what you see in 2Us: more card options thanks to having a full-height interior, more hot-swap bays up front, and often room for a third, redundant power supply for an N+1configuration. (This means the needed number of PSUs, N, plus one spare for redundancy. Usually, these redundant power supplies will be hot-swappable.) 4U and 5U cases often take a step back in storage capacity because, as we saw with Intel’s SC5400, such cases often begin life as pedestals. The exception is with 4U/5U servers that host 4P motherboards. These are almost exclusively rackmount systems and trade much of their storage real estate for additional fans to keep those extra chips cooled.

We expect Intel to hammer home the modularity message in ever-varied ways over the next couple of years. The opening shot was blades, but pedestals can’t be far behind. The more modularity comes down in cost and proves itself on ROI, the more we’ll see “real” servers mount an even more persuasive argument against pseudo-server desktops.

[ READY FOR MORE? ]

We’ve covered a lot of ground so far, but there’s still more to come. Energy consumption is a hot subject in the server market, but how much of a consideration is it for your SMB clients, and what’s the best way for your boxes to go green? Where does software fit into your product equation? What changes will you need to make in your organization as you expand your focus from desktops to include servers? Perhaps most important of all, now that you’re up to speed on the foundations for making solid servers, what are the easiest market targets to aim for? Don’t miss out. Head to reselleradvocate.com for Part 2 of our deep dive into designing best-of-breed servers.

Page