Server Architectures: Dense, and Getting Denser


In server architectures, scalability is still king. But when it comes to achieving the levels required for advanced cloud computing, Big Data and increasing mobile traffic, the need for scalability must be tempered by the need to keep costs down and hardware footprints small.

For these reasons, it’s getting near impossible to talk about scalability without delving into server density. The overriding question: How dense can you go before you start to counter the performance advantages and operation efficiencies that highly scalable architectures are supposed to provide?

Clearly, the enterprise industry does not feel it is anywhere near that point yet. In fact, while demand for server configurations is down nearly across the board, the one bright spot is what IDC identifies as “Density Optimized Servers.” The sector grew nearly by half in the first quarter to 2013, compared to the same period a year ago to $753 million, with unit shipments surging some 40 percent to just under 215,000. While still only a fraction of the overall server market, it nevertheless provides a strong indication that as new cloud infrastructure comes on line, and established organizations seek to convert legacy architectures into private clouds, high density will be one of the hallmarks of the new data environment.

Clearly, the server market is in the midst of a significant transition, and it isn’t entirely clear whether some of the leading server manufacturers are fully prepared. Dell, at least, seems to be in pretty good shape, having garnered nearly half of the density optimized market with its PowerEdge-C family, but at the same time there is a growing cadre of smaller manufacturers who smell blood in the water.

Taiwan’s MiTAC, for example, is the first out of the gate with a 64-bit ARMv8 machine. Targeted at enterprise and cloud-scale workloads, the 4U 7-Star system sports 18 server cards, each of which holds an 8-core ARM from AppliedMicro, as well as a pair of DDR3 DIMM slots and room for two solid state or hard disk drives. As well, each card gets its own 10G SFP+ port and a 1 GbE port.

Meanwhile, a company called Aaeon is ramping up a new ARM solution of its own, turning toward Calxeda to provide its EnergyCore processors for the Indus storage-optimized server platform. The system is intended to work with block, file and object applications, utilizing twin 10 Gbps links to handle traffic from two compute notes and up to 10 3.5-inch HDDs, all housed in a 1U chassis. A key element in the design is the fact that the EnergyCore platform supports application deployment though a standard ARM-based Linux release.

ARM is by no means dominating the density market, however, at least not yet. Supermicro is using the new Intel Haswell processors in its uni-processor (UP) platform. Using the 22 nm, Tri-Gate structure of the Haswell (also known as the Xeon E3-1200 V3), Supermicro is planning on 8-, 12- and 24-node configurations featuring hot-swappable components, 32 GB memory, and support for either two 3.5-inch or four 2.5-inch HDDs in a 3U design. The company is also planning multiple 1U devices optimized around general workhorse processing, networking and other functions.

Density in server architectures represents more than just a bid to increase processing capabilities within existing data center confines. It also enables organizations to ramp up efficiencies to deal with the twin challenges of more data and smaller budgets.

And as data becomes increasingly characterized by either the large-volume, large-packet loads of Big Data or the large-volume, small-packet needs of the Web, highly dense configurations seem to be the best way to fulfill both needs.