X
Tech

A reason to refresh

Intel did more than just introduce a faster server processor today with the introduction of the Xeon 5500: it enabled a greater level of differentiation to its server and storage vendor partners that ultimately will result in a broader set of choices and better ones for enterprise infrastructure & operations professionals. While the performance improvements of the 5500 in themselves are impressive, there is just as much to like in the new memory and I/O architectures and power efficiency.
Written by James Staten, Contributor

Intel did more than just introduce a faster server processor today with the introduction of the Xeon 5500: it enabled a greater level of differentiation to its server and storage vendor partners that ultimately will result in a broader set of choices and better ones for enterprise infrastructure & operations professionals. While the performance improvements of the 5500 in themselves are impressive, there is just as much to like in the new memory and I/O architectures and power efficiency.

The new memory architecture triples bandwidth over the 5400 and brings back DDR3 allowing up to 18 DIMMs per CPU. This lets customers reach much higher memory configurations at a lower cost. While you have to add memory three DIMMs at a time, 36 GBs per socket is now achievable with low cost 2GB DIMMs. This is a significant boon to server virtualization where memory is typically the first resource to be fully utilized. Cisco is taking this capacity even higher in its UCS blade servers.

On the network bandwidth front the new QuickPath I&O architecture delivers significantly more bandwidth to each CPU -- over 25GB/s per lane -- enabling massive I/O capacity for virtualized workloads. These benefits, in combination with improvements in CPU design, bring value to hyperthreading on the Intel architecture for the first time -- another significant value for virtualization.

While certainly all these benefits can be leveraged for high performance applications and will drive upgrades that increase the productivity of these applications, perhaps the biggest reason to consider upgrading sooner to the 5500 is for the faster return on investment you can realize by achieving higher levels of server consolidation. Inquiries with many Forrester enterprise clients have revealed that the average utilization of virtualized environments lays around 40 percent average with peaks around 60 percent. While a big improvement over 5 to 12 percent consumption of physically deployed workloads on average, we can certainly do better -- and should to maximize the benefits of our virtualization investments. Forrester recommends clients strive for 60 percent utilization with peaks of around 80 percent. Read our recommendations on how best to achieve this.

The new servers coming to market based on the Xeon 5500 will help you achieve much higher consolidation while reaching for this utilization target but it won’t be as much of a one size fits all proposition as it has been with prior Xeon servers. Due to the new architecture, server manufacturers have a lot more options for creating differentiated solutions that help IT Ops professionals pick the right systems that best meet their organization’s objectives. We are already seeing evidence of this through the new servers announced by Dell, HP, IBM, Sun, and others that provide a much wider range of configuration options and densities. Plus the Xeon 5500 finally addresses SMP applications with an architecture that allows for true 8-way server designs.

But for power-starved data centers, or those leveraging colocation facilities perhaps the biggest advantage the Xeon 5500 brings is dramatically better power efficiency. Intel got it right by making these capabilities dynamic as well. While server administrators always like to have the knobs to turn, deriving the most power efficiency from the architecture should be a high level decision, not one where you have to dive into a collection of microknobs to get there. Just as HP and IBM’s power capping software lets you set top line power consumption thresholds you should not have to determine whether you want two cores turned off, for overclocking to happen or to manage the power behavior of memory controllers, I/O controllers or other subcomponents. Let the system make the appropriate microadjustments based on the needs and behavior of the applications they are executing and your top line guidance.

What this means for IT Ops professionals is the ability to set your power thresholds per rack and maximize the compute per watt within that threshold. Thus for IT consolidation, strive to maximize the utilization of not just the resources within each new server you purchase but the power envelope of your data center. The performance per watt results that are now achievable in the new systems leveraging Xeon 5500 CPUs are a strong incentive to accelerate the refresh of your server systems and achieving a much lower on-going infrastructure cost envelope.

While it may sound counterintuitive in a cash crunch economy like today’s to be recommending the purchase (and accelerated purchasing) of new IT equipment, remember that for every less efficient server you remove from your environment you end a support contract, free up admin time, recoup power and rack capacity, reduce the consumption of OS licenses and thus bring down overall TCO. If you have achieved 10:1 server consolidation using existing systems and can get to just 20:1 with the new systems, the relative cost of operations savings alone may fund the capital investment all within the same year. It’s time to accelerate your consolidation efforts.

Editorial standards