To hear some people tell it, there is no such thing as too dense when it comes to the server room. A more compact footprint, and a greater number of VMs, can only be a net gain in the data center, both in terms of processing capability and energy efficiency.
There is nothing wrong with a dense server rack, of course, and it seems that there is still plenty of room to pack things even tighter as the struggle to keep up with social networking and Big Data continues. But at some point, this well must run dry, won’t it?
The server industry certainly hopes not, considering dense platforms are one of the few bright spots in an otherwise dismal market. According to IDC, worldwide server sales slid another 6.2 percent in the fourth quarter to just shy of $12 billion, reflecting a 1.2 percent drop in shipments. However, machines in the category that IDC calls “density optimized servers” saw growth of 26.6 percent to about $735 million. At only about 200,000 units, it is relatively small compared to the wider market, but growth is growth, and if density is what the market wants, you can bet the server vendors will be more than willing to oblige.
IBM, for example, recently unveiled a new series of Flex machines that the company describes as a “double-density” platform intended to support software-defined architectures for high-volume workloads. The Flex System x222 is in fact two independent servers tucked onto a standard blade platform, which means up to 28 machines can be packed into a single 10 U chassis capable of supporting up to 2,800 virtual machines. You also get up to 384 GB of DDR3 memory and twin 10 GbE ports per node and an additional I/O connector for Fibre Channel or InfiniBand service.
The use of both on-server memory and high-speed connectivity is crucial to highly dense server configurations, considering that ramped-up processing is of little use if it can’t access stored data in a timely manner. Of course, the more that can be stored close to the server, the less that has to traverse complicated network architecture, which is why the server industry is looking forward to the new DDR4 modules hitting the channel. Samsung, for one, is starting mass production of next-generation 4 GB DDR4 units built on a 20 nm process. Not only do they push memory modules into the 32 GB range, but they boost transmission speeds to 2.6 Gbps while cutting power consumption some 30 percent over DDR3.
Increased density in the server room affects more than just storage, however. While it’s true that higher density configurations increase resource utilization and energy efficiency, these gains can be diminished if not coupled with an updated power and cooling infrastructure. As FieldView Solutions’ Sev Onyschkevysh noted recently, dense systems run hotter, and cooling systems designed for entire rooms are not the best when it comes to servicing individual units – further evidence that the data center is an organic entity, not merely a collection of devices.
High-density architectures in the data center are gaining popularity in large part because the data center itself is changing – from the enterprise-based entity of yesterday to the cloud-facing architectures of tomorrow. In this dynamic, modular scale-out infrastructure is still king, as long as it can be contained in the smallest space possible.