Dawn Moore, GM Networking Division
Data center application performance today uses balanced system performance based on a combination of CPU power, faster storage and high-throughput networks; upgrading just one of these elements will not maximize your data center performance.
This wasn’t always the case. In years past, some IT managers could postpone network upgrades because slow storage would limit overall system performance. But now, with much faster solid state-state drives (SSDs), the performance bottleneck has shifted from the hard drive to the network.
This means that in today’s IT environment—with hyperscale data centers and virtualized servers—it’s crucial that upgrading to the latest technology, like faster SSDs or 10/40GbE, be viewed from a comprehensive systems viewpoint.
Certainly, upgrading to a server with a new Intel® Xeon® Processor E5-2600 v3 CPU will provide improved performance. Similarly, swapping out a hard drive for an SSD or upgrading from 1GbE to 10GbE will improve performance.
Two recent whitepapers highlight how maximum performance depends on the interconnected nature of these systems. If the entire system isn’t upgraded, then the data center doesn’t get the best return from a new server investment.
The first paper* discusses the improvements in raw performance that can be seen in a complete upgrade. For example, when an older server with SATA SSDs and a single 10GbE NIC was replaced with a new Intel® Xeon® processor E5-2695 v3 based server, a PCIe SSD, and four 10GbE ports, the new system delivered 54% more transactions per minute and 42.4% more throughput, as well as much faster response times in these tests.
What can be done with this raw performance increase? The other whitepaper** answers that question by researching the increase in the number of virtual machines supported by an upgraded system.
With SDN in the data center, data center managers can facilitate the ramp up of new virtual machines (VMs) automatically as user needs grow. In the case illustrated in this paper, it was the ability to automatically spin up a VM and a new instance of Microsoft Exchange to support new email users. With all of this automation, the last thing that’s needed is for the infrastructure to restrict that flexibility.
In this example, a Dell PowerEdge R720 server replaced an older Dell PowerEdge R710 server-storage solution. These new systems featured the latest Intel® Xeon® processor, new operating system, SSD storage and Intel® Ethernet CNA X520 (10GbE) adapters. When the tests were finished, the new system supported 4.5 times more VMs than the previous system.
What is interesting to me is that the researchers measured the performance increase for each part of the upgrade—which really illustrates the point that these upgrades need to done comprehensively.
In this test, when the researchers upgraded just the CPU and the OS, they saw performance increase 275 percent. Not bad. But when they added the higher-performance SSDs to the new CPU and OS that resulted in a 325 percent improvement. And finally, when they added the new network adapters, overall VM density improvement climbed 450 percent compared to the original base system.
More details on both of these examples are available in the white papers referenced below.
When it’s time to invest in new servers, take a look at the rest of your system, which includes your Ethernet and storage sub-system, and think about the combination that will give you the best return on your investment.