Data center design, Itanium roadmap, power management

Rory McInerney discusses power management, the impact of high end processors on data center design. The future of Itanium. What facilities should ask IT prior to a server refresh. Power and heat management. And what’s happening inside Intel’s own data centers.

Energy Management and energy Performance: 
In 2006 your data center might have had Intel’s dual core servers. Replace those today with the E7 and we can reduce 18 racks to one and save 93% of energy and deliver a massive consolidation and lower footprint. We’re on a vector of power manager with products like data center manager and node manager. The server is still consuming the most energy within the data center. 
What about Energy management? 
You’ve bought this hardware. The question is how can I do more work on it. I’ve designed my data center for peak loading and I’ve bought the hardware and that’s were virtualization comes in. As people virtualize their average utilization continues to go up. And that’s where people see the value in buying faster and more powerful servers, because they are getting more out of it. Your value in servers goes up if you can operate at higher utilization more often, then the more work gets done.  And that's not from the perspective of using more power, your average energy use over time will be better and that means you don’t need as much infrastructure. Or it can mean you can put more load on so you can expand your business without having to expand your data center.  

We’re giving the end user much more visibity into what’s really happening at the CPU and what’s happening over time. We do a lot of turning off things or powering off things. Just using the hardware that’s needed for a particular application or making the hardware more application aware. We are innovating on that space and giving more visibility 

Common infrastructure and scaling 
There are still interpretations – we’re in a dynamic environment. How do you allocate resources in a dynamic environment? Do you do it in at the rack level, at the server level or do you go into the core.  You have to go the [hardware] providers to see if they really go down the actual core and some of them will want to provide that level of granularity.
 
We’re providing 10 cores in this product. For the E7 on 10 cores that is a little different from doing dynamic provisioning on 2 cores. 

How do become efficient?

There are two ways to get efficient. Take inefficient equipment out and put efficient hardware in, and maybe less of it, because you can consolidate. And once you have that infrastructure in provide visibility to how that is being used such that you make smarter decisions dynamically.
Itanium – roadmap:
Following the Xeon E7 announcement will come Poulson – for release in 1st half of 12. This is part of integrated roadmap. We’ve an Itanium roadmap and an E7 roadmap – and they are not exactly timed the same. The rates which systems get refreshed is different in those two markets. Our Itanium roadmap stretches as far as our Xeon roadmap. Poulsion is at 32nm – the E7 is also at 32nm and we haven’t disclosed anything about our future products. 

A typical data center scenario – if I’m moving to a Westmere E7 processor – what should I thinking about in terms of my energy efficiency. What kind of energy savings can I see if I am deploying these latest processors?
You have to compare to what? We have a power envelope that we design these CPUs for. We collaborate with hp, IBM and Dell on this power envelope. We have a power envelope that we design these CPUs for. Our typical Xeon, DP2 socket – is probably operating on a 130W CPU envelope. There are lower performance nodes, depending on the data center you are trying to design. Below the Xeon 130W, we have 90W products and 60W products.
 
What typically happens is that you look at how much memory is being deployed. When you go to lower power products, not as much memory, not as many i/o channels, so you are becoming targeted and less scalable in terms of the applications. So within the Xeon family you can see products with half the power within that same platform family. We see those trends staying, so when we look at our design points over time, what those numbers will be, in terms of Intel design points, will stay. The amount of volume that ships will change [in relation to the power performance] and that’s something we tend to see depending on performance per watt. What is my performance per unit of energy for my application? 
The Atom family of Intel products– really low wattage – for deployments that are not performance centric but is extremely power sensitive. 

At the CPU level, we have solutions at ultra low power all the way to Itaniums at 170W. So the question for the facilities guys [to ask the IT guys] is which one of those you are looking at deployiong and what are you replacing.

And I would say within the Intel family, you are largely fitting in to the same ecosystem you already have, we’re very cogniscent of making sure you don’t have to do something special, you don’t have to do special cooling over and above what you already have.