You know the requirements your new physical server must meet, but how do you align what you need with what's available to you? You must start by understanding server options -- as well as the long list of buzzwords -- to determine the features that are most important to your environment. In other words, let's get through all of the noise.
As explained in "Buying a server? How midmarket CIOs can determine their needs," there are a number of different types, styles and models of physical servers. The most commonly deployed servers are the small, floor-standing pedestals, towers, under-desk or desktop servers, as well as 1U and 2U rack servers. But understanding server options that are best suited to your business requires you to consider a number of very specific areas.
Processing GHz, sockets, cores and threads
A processor or chip sits in the core of every server, taking on the actual computing work of accessing memory and I/O devices.
As with understanding server options, there are a number of processor choices to wrap your head around: 32- or 64-bit, x86 open or proprietary, single or multi-core and various forms of intelligent power management schema, memory and I/O options. Beyond the basics, some processors also provide some advanced functionality, capable of automatically turning cores on or off (depending on workload demands via application or operating system software tools) as a way to save energy during idle periods, or increasing or decreasing clock speed to either boost performance or slow it to a leisurely pace.
A server processor can contain one or more cores, meaning a single socket could have a dual or quad processor supporting one or more active threads. A server with a single socket, single core and single thread can execute only one instruction or operation at a time. On the other hand, a dual socket and quad core with one thread could, software permitting, execute eight instructions simultaneously without constraint. Likewise, a quad socket, quad core would enable 16 instructions, allowing a hypervisor and operating system to potentially allocate those resources to virtual CPUs for application performance load balancing.
If you need more performance or processing power than a single server can provide, cluster across server blades, individual rack mounts or floor-model servers. Assuming that operating systems, hypervisors and applications can utilize concurrent threads, cores and sockets, this increased density can address various performance and application needs. However, if your current applications, operating systems or hypervisors are not able to fully utilize all of those features, don't be up sold.
Also keep in mind that licensing models have changed. Some applications and software are no longer licensed by the size of the server alone, but based on the number of cores and sockets. Perform the necessary due diligence to ensure that you are meeting the licensing requirements when using multi-socket, multi-core processors.
Understanding server memory
Computers rely on memory. Server memory, which includes external disk storage, is used for storing operating system software and all the associated tools, utilities, application programs and data.
Main memory or RAM, also known as dynamic RAM (DRAM) chips, is packaged in different ways, with a common form being dual inline memory modules. DRAM memory access speed is referred to in terms of older DDR2 (667 MHz) or newer DDR3 (1333 MHz). RAM main memory on a server is the fastest form of memory, second only to internal processor or chip-based registers -- L1, L2 or local memory.
In general, more memory is better; however, the speed of the memory is also very important. Different versions and implementations of virtualization solutions support various memory configurations and limits. Check with specific vendors' compatibility lists for supported configurations and memory requirements. Also check with vendors for supported configurations of 32-bit and 64-bit processors, single-core, dual-core, quad-core or eight-way processors, along with I/O cards and drivers for networking and storage devices.
Networking with your storage and your users
Take a look at what functionality is built into the server or provided on server blades for general-purpose networking along with attachment of disk storage. What is there in terms of 10 Gb Ethernet (10 GbE), and how many ports as well as 3G (3Gb) or 6G (6Gb) Serial-Attached SCSI (SAS) for disk storage attachment (internal or external), along with serial, video and USB ports? Also look at expansion capabilities for additional mezzanine cards for blade servers, or PCI-E cards for networking, storage and other peripherals.
PCI SIG Multi-Root IO Virtualization (MR-IOV), a relatively new and emerging feature for servers, enables advanced connectivity, including adapter sharing. MR-IOV will enable multiple, physically separate adjacent servers to share a PCI-E adapter card, allowing the virtualization of servers that otherwise could not be consolidated. MR-IOV can also boost scaling capabilities beyond normal physical limits in high-density servers by placing adapter cards in shared external expansion slots.
Greg Schulz is founder of The Server and StorageIO Group, an independent IT industry advisory and consultancy firm based in Stillwater, Minn. He is also author of the books The Green and Virtual Data Center (CRC) and Resilient Storage Networks (Elsevier). Follow him on Twitter at @storageio.
This was first published in July 2010