Supermicro Deserves Some Respect

I’ve long thought that Super Micro Computer, or Supermicro, was the Rodney Dangerfield of server vendors. After years of delivering solid, if not especially exciting, servers, the company still don’t get no respect from most customers, who'd rather buy their servers from HP, IBM or Dell. Supermicro may be worth a second look, as its products have evolved from generic, one- and two-processor tower and rack-mount servers, and may be just the cost-effective solution to one or more of your server problems.

OEMs have long relied on Supermicro servers to house their software. Many of the spam filters, virtual tape libraries and iSCSI arrays that have come through the NetworkComputing Real World/DeepStorage Lab over the years have been based on Supermicro hardware. Smart appliance vendors look to minimize their warranty service costs, so I count long-term success in the OEM market as a sign of hardware reliability.

For the past few years, Supermicro has been the go-to player for when the standard 1U and 2U packages just don’t fit the need. Need high server density while retaining the I/O flexibility of standard PCIe slots? The twin series doubles server density by putting two servers, with the usual set of memory and PCIe slots, in a 1U or 2U chassis. The company can even stick four dual Nehalems, each with one PCIe slot, or eight Atom servers in a 2U chassis. The MicroCloud squeezes eight single-processor Xeon servers in a 3U space. The chassis provides drive slots and power without the management or I/O in a typical blade chassis.

Want to build a high-density storage server with Openfiler, NexentaStor or the like? Supermicro has a server chassis that loads drives in from the front and the back holding up to 45 LFF 3.5-inch drives or 88 SFF 2.5-inch drives.

If your software supports clustering, the company's dual server SBB (Storage Bridge Bay) system puts two dual Nehalem servers in a 3U cabinet with two 10-Gbps Ethernet links between them and 16 SAS or SATA drives over a shared backplane. Look closely at most second-tier vendors' dual controller systems, and you’ll see an SBB system.

Until I spoke with Supermicro, I thought of the company as innovating primarily in mechanical engineering while taking the Intel or AMD chip set and assembling pretty standard motherboards. As we walked thought the product line, I started noticing systems designed for GPU computing with multiple Nvidia Tesla GPUs and, most impressively, an eight-way Xeon 7500 system that can take 2 Tbytesof RAM. That’s a bigger server than you’ll find in the Dell product line, and there is no Intel reference design.

The big boys still have some hardware tricks up their sleeves that Supermicro can’t match, like IBM’s lightpath LEDs on the motherboard that point out the failed component to your server techs. the Max 5 massive shared memory expansion device or Cicso’s expanded memory feature for UCS blades.

One problem many users have is that Supermicro still operates as an OEM supplier. While the other vendors sell their servers complete with processors, memory disk drives and the like, Supermicro just sells the parts it makes--primarily, the chassis and motherboard.

That means those willing to do a little assembly--like me, when we needed more servers in the lab recently--can save a few hundred to a few thousand dollars per server because they don’t have to pay the vendor’s markup on commodity parts such as memory and disk drives.

It also makes those less-geeky people uncomfortable, as they have to assemble their systems and choose parts. HP may send you the memory, disk drives and additional processor in separate boxes, but at least they all say HP on them. As a result, Supermicro systems are seen more in technology companies and universities than banks and brokerages, though some resellers have seen they can assemble and test systems, and still sell them at a competitive price.

Do I think every organization should put Supermicro on its server short list? No. But I think more self-reliant users can save some money and still have reliable, high-performance systems in their data centers or home labs.