Cisco Blogs


Cisco Blog > Data Center and Cloud

VMware’sVMMark Benchmark: Number One Position for UCS on Intel Xeon for 8 Core 2 Socket Systems

February 4, 2010
at 12:00 pm PST

The importance of reviewing data for Virtualization 
by Harris Sussman,  Cisco Data Center Solutions -- Unified Computing System

 

When buying a car, you have a choice to do research, ranging from perusing the mfg’s brochure, Consumer Report or actually driving a vehicle. Choosing a server vendor is similar, in that an IT Mgr. needs to assure the server meets the criteria for their business objectives.

 For most IT buyers, purchasing decisions are not trivial, and each organization applies their own philosophies. As IT staffs embark on new virtualization projects, the aim is to reduce cost, increase business agility, and reduce complexity. There are a plethora of tools and industry benchmarks available, but when it comes to virtualized environments, it’s critical organizations get this decision right.

While most Hypervisor vendors have adequate benchmarks for their respective products, VMware’s VMmark is still perceived as the gold standard, The which combines 6 of the most common DC workloads running within a unit of work referred to as a tile. This methodology is still the most sought after.

VMware’s VMMark benchmark is one of the most active benchmark sites where vendors are constantly trying to improve their results. Just recently, Cisco published their latest UCS results http://www.vmware.com/products/vmmark/results.html to regain the number one position using the latest Intel Xeon processor for 8 core 2 socket systems. While bragging rights are important for the vendor, customers rely in this info for buying decisions.

Performance benchmarks are an important data point, but in the absence of a standard virtualization benchmark businesses must do the due diligence necessary to ensure they  choose the right Hypervisor. Chris Wolf, A Burton Group analyst posted a nice blog about the need for SPECvirt  now http://www.chriswolf.com/?p=303.

 

To learn more about the UCS family visit here - More benchmarks on UCS

In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.

9 Comments.


  1. Hmmm, I just checked, and the highest 2-socket 8core result is for Fujitsu RX 300 S5 with a result of 25.16@17 tiles compared to the Cisco result of 25.06@17, am I missing something?

       0 likes

  2. Omar Sultan

    Nik:The Fujitsu system is based on the W5580 workstation CPU which is not supported for servers.Regards,Omar

       0 likes

  3. Omar Sultan

    Harris:Good post–its useful to note that, a year ago, the top benchmark score was 11.28, so we have seen the top benchmark more than double in a little over a year.Omar

       0 likes

  4. I’m sorry, but using the W5580 version of the Xeon doesn’t make the Fujitsu result in anyway ineligible. The W5590 has exactly the same feature set in terms of support for things like ECC memory. The only difference is the higher clock speed and thermal envelope. You can make an arguement that the Cisco box is more power efficient given that it delivers almost the same performance. But the claim made here is that this is the fastest server result, and the Fujitsu RX 300 is most definitely a server, just take a look at it…http://ts.fujitsu.com/products/standard_servers/rack/primergy_rx300s5.html

       0 likes

  5. Omar Sultan

    Nik:The W5580 is not certified by Intel for server operating environments–see the note on this page: http://www.intel.com:80/cd/channel/reseller/asmo-na/eng/products/server/processors/5500/feature/index.htm. Hence, we felt it was reasonable to exclude it. I am guessing most folks would not want to take the chance, given as you note, the potential performance gain is minimal.Omar

       0 likes

  6. But it’s not a technical issue, it’s down to Intel thinking that a processor with such a high TDP isn’t suitable for server applications. If you have a server chassis with the power and cooling budget to support it’s a perfectly valid choice. The main issue I have here is that you had a perfectly good story without arbitrarily redefining who qualified to be part of the benchmark.If you’re interested, you can see my thoughts on this over at http://dcsblog.burtongroup.com/data_center_strategies/

       0 likes

  7. Omar Sultan

    Nik:I guess we will just have to agree to disagree on this one. Intel has said they would not support the 5590 in server applications, so it more than just marketing positioning on Intel’s part, so we felt it was reasonable to exclude it based on my earlier comment. That being said, you point about VMmark perhaps being CPU bound is a fair assessment. Folk that follow this blog know what a big fan of synthetic testing I am.Regards,Omar

       0 likes

  8. With reepst to CPU performance, my point was that VMmark isn’t CPU bound on current gen Nehalem chips, otherwise there would be significant gap between results on 2.93GHz and on 3.3GHz. Since there isn’t I’m guessing that we are hitting memory bandwidth or I/O bottlenecks at the moment.

       0 likes

  9. I just love the memory extenders on the UCS blades, that really makes the difference in vmware use cases.

       0 likes