Virtualization Projects, Misguided focus on CPU vs. IO, Infiniband vs. 10GbE

Voltaire had a press announcement about how IT executives lack of the green (money) to go green.

Voltaire Survey Shows IT Executives See Greening of Data Center as Mission-Critical, But Lack “Green” To Go Green

Voltaire Develops “50-50-300” Pledge and Efficiency Calculator in Response to Findings

BILLERICA, Mass. and HERZLIYA, Israel – September 3, 2008 – A recent survey by Voltaire Ltd. (NASDAQ: VOLT) found that, although CIOs and senior IT executives overwhelmingly believe that a green data center will become mission-critical, many lack the “green” to go green.

Nearly 90 percent of executives surveyed said they believe that greening their data centers will be crucial to meeting their companies’ business objectives in 2009, and 57 percent said they believe going green gives them a competitive advantage. Yet, 76 percent do not have a committed budget for a greening policy. The survey queried CIOs, CTOs, and senior IT executives who attended the 2008 MIT Sloan CIO Symposium.

In response to these findings, Voltaire developed a “50-50-300 Pledge,” which states that IT executives, working with the company to deploy a Voltaire unified fabric, can save 50 percent on power/cooling related to server interconnections and 50 percent on hardware allocation/usage, while delivering up to a 300 percent increase in application performance. Voltaire has also developed an Efficiency Calculator ( to help IT executives estimate their network energy and cost savings and justify the investment.

“It appears from these findings that senior IT management is still in the planning phases, and they will need to prioritize funding for these important greening initiatives,” said Patrick Guay, Executive Vice President Global Sales and General Manager of Voltaire, Inc., the U.S. subsidiary of Voltaire Ltd. “We help them with the analysis and data collection in order to validate the savings they will achieve. Our numbers show that for enterprises, the return on a green data center fabric infrastructure using currently available technology is in the millions of dollars. For example, a Fortune 500 company with five data centers worldwide, and 3,000 servers per data center, can save approximately $7,400,000 per year.”

I was able to get time with Patrck Guay  on Sept 5 to discuss more about Voltaire's solution. What has me interested in Voltaire's is their strategy of going after the early adopter market of areas like supercomputing and high performance computing where performance per watt is becoming a standard practice. As users have needed faster IO, evaluating technologies like 10Gb Ethernet, Voltaire has been able to get Infiniband based solutions in as better performance per watt solution. I asked Patrick if he saw much interest in his products in the dev/test/quality server labs in enterprises, and he said this area is another early adopter of Voltaire's product with a 24 port switch including full Infiniband network support in an entry switch.

High Performance, Low Latency InfiniBand Switching for Small-to-Medium-Sized Clusters and Grids
isr 9024, Linux, low latency, Massachusetts, non-blocking switch, supercomputer The Voltaire Grid Switch 9024 is a high performance, low latency, fully non-blocking switch for high performance computing (HPC) clusters and grids. Offering available bandwidth of up to 960 Gbps, the Grid Switch 9024 is a cost-effective alternative to proprietary interconnect technologies. With twenty-four 20 or 10 Gbps ports in a 1U chassis, the standards-based Grid Switch 9024 delivers high bandwidth and low latency at affordable pricing. Using the Grid Switch 9024, you can build high performance clusters and grids that scale from several to tens of nodes.

    Key Features
  • 10-20 Gbps performance for clusters and grids
  • Ultra-low latency: under 140 nanoseconds
  • Available bandwidth of up to 960 Gbps
  • Built-in high availability
  • Ideal for scientific, commercial HPC and enterprise applications
  • Architected to provide high MTBF
  • Powerful CPU to allow management of fabrics, as well as device management capabilities

Here is Voltaire's 50-50-300 pledge.

Voltaire is the only company today that can deliver a unified fabric based on InfiniBand, which uses 10x less power per port than
10 Gig E
. Unified fabrics provide seamless, high performance networking services between InfiniBand fabrics, Fibre Channel SANs and Ethernet LANs over a single high performance fabric with multiple virtual interfaces replacing actual physical adapters.

Most of the focus on Greening the data center is on the CPU, getting the most performance per watt out of each processor. The problem is this leads many to focus on a high CPU utilization as a measure of efficiency. You can imagine how some IT departments are showing off their CPU utilization numbers as proof they are more efficient.  But, as many of us have experienced in a bad application, driver, or technology like indexing that hogs the processor on your laptop - there is a flaw of a narrow-minded focus on CPU utilization.

Most server HW is IO bound, by memory, disk, network, and CPU IO to these devices.  As HP just announced its BL495C blade for virtualization, HP focused on 3 issues to tout its strengths - more DIMMs, SSD, and 10GbE - all IO issues. The blade itself is only 2 quadcore AMD processors with claims of supporting up to 32 VMs.

Which brings up an interesting side by side comparison.  HP compared itself vs. Dell and IBM. What I would like to see is a Sun based Infiniband (Voltaire) blade comparison vs. HP's BL495C & 10 GbE.

  1. 10 GbE vs. Infiniband.
  2. HP's AMD implementation vs. Sun's AMD implementation. 

What is more efficient? 

And that brings up the problem of how do you get numbers you believe and trust.

Which probably explains why Voltaire has seen the need to pledge its claims as we have all learned to filter the greenwashing.

Voltaire has a white paper here.

White Paper

Reducing Data Center Energy Costs up to 50% By Consolidating and Virtualizing your Network

This white paper describes how combining data center networks into one "fabric" will have a significant impact on o