Summarizes the work Cisco IT has been doing to compare the benefits of building such data centers with the Cisco Unified Computing System (UCS) and Unified Fabric versus “traditional” data centers.
4. Case Study – 1 MW Data CenterNew design – impact on power by using Unified Fabric 47% fewer racks Savings Of ~5,500 cables ~30% more power for servers
5. Cisco IT’s Case Study– Summary Density Efficiencies – 47% fewer racks 40% Savings from cabling Almost 4x VMs per KW 12,000 to 28,000 VMs – In the same Data Center! Notes: Assumes pre-UCS average V2P ratio of 10 to 1 and post UCS average ratio of 20 to 1 due to the memory expansion technology. Unified Fabric efficiency gains result from power optimization. UCS efficiency gains result from additional power benefits of UCS.
6. Cisco IT Case Study Summary More Capacity to grow “The Business” Elasticity to align DC Capacity with Business Demands “Pay as you Grow” drives lower operating expenses Foundation for ITaaS Model Energy efficient deliver of Services Building the Cloud of tomorrow today
7. Cisco on Cisco Data Center Experience www.cisco.com/go/virtualdatacenter
Notes de l'éditeur
This is what we’re doing right now. It’s a real life story.Regarding the Physical servers: With 8 cores per server, this translates into 5,760 cores out of 720 physical servers. (a measure of capacity)As this is a Greenfield situation, you may be wondering how we’re implementing this in our legacy data center. It’s currently in the planning stages. We think that power is an issue, but because UCS will give us these VM improvements, we’ll get a VM power increase.Note that this is FOR OUR WORKLOAD, but the trend will be similar to what you have. There are sweet spots in memory configuration. Technically it’s 8 core…. Which will allow us to get to 30 (?) vms on one server.Physical server count: range “ UCS has advantages for certain types of applications that are memory-bound. If your apps are memory bound, and you take the upper limit of 1500 servers that are avail, then the memory optimization will give you approx 20 VMs if your original workload was 10vms. If you’re more conservative, then take only 10. These are conservative numbers.Important to note: this is in the same size data center!The virtual machine count (bottom) is significant. You can get at least 10 VMs per server. Depending. The Vm count depends upon the size of the CPU and the memory.Using unified and some other blade server, you save power, which is translated into more physical servers. We’re putting more power to the servers so we can have more servers. In our case, we were wasting power in each rack. As we made the density more, in aggregate, so there’s less unused power on the floor. From 720 to 100 is because we’re not .We’re building a new environment in Amsterdam, planning for such a pod there. 1 QTY Logical Component Estimated Power Requirement (kW) 2 NetAPP - 3070 (VTL700) 0.826 Single Pod Summary0.83QTY per Zone 0 QTY Logical Component Estimated Power Requirement (kW) 2 MDS -9513 0 Single Pod Summary0.00QTY SDF - Storage 1 QTY Logical Component Estimated Power Requirement (kW) 8 MDF - 9513 20.8 Single SDF Summary20.80QTY DMZ 1 QTY Logical Component Estimated Power Requirement (kW) 4 NetAPP - 3070 (VTL700) 1.652 4 MDF - 9513 10.4 Single Pod Summary12.05QTY Disk Rows 1 QTY Logical Component Estimated Power Requirement (kW) 4 EMC DMX3 Tier 1 104.8128 4 EMC DMX3 Tier 2 102.72 0 EMC DMX3 Tier 3 0 5 NetAPP VTL/Filer w/ 12 Shelves 20.305 4 Quantum PX 720 6.4 Single Pod Summary234.24