In order to achieve greater efficiency and lower power in their data center, a North American insurance provider began to evaluate new solutions to combat problems within their data center. The first problem identified was poor port utilization across their network due to their top of rack switching environment. The data center server cabinets in the initial constructed phase can only house 12 to 14 servers and the top of rack switches dedicated to each server cabinet had very poor port utilization. In fact across the entire date center their port utilization was approximately 19%.
The second problem was stranded power and power ports, which is common in many data centers. Stranded power is defined as power allocated but unused or unusable and therefore depletes capacity available for other power consuming resources such as switches, servers, etc. In addition, there was a wish list to enhance cable management and overall aesthetics in the data center. The insurance company began evaluating options to solve these problems.
Based on a recent webcast and white paper from Siemon, the company sought an evaluation to determine if like savings could be realized in their environment. In a white paper from the CCCA (Communications Cabling and Connectivity Association) savings were identified in power and equipment costs using 10GBASE-T and end of row switching instead of top of rack switching as currently deployed. As the data center was about to launch another phase with 10 additional pods (two rows of cabinets per pod) it seemed a good time to determine what savings could be achieved for the new build with an eye to remediate the current environment in the near future.
In an evaluation comparing a top of rack switching architecture to an end of row architecture utilizing 10GBASE-T switches, significant savings were identified in switch purchases, cabling, power, and maintenance costs. In fact, the costs for the switching environment decreased from $17,623,040 MSRP in a top of rack configuration to $7,895,625 using BASE-T switches at the end of a row (excluding software and maintenance costs) for a total identified savings of $9,727,411. In addition, fewer switches result in lower OPEX costs for power and maintenance. In fact, power consumption in this model would be cut nearly in half.
Once the switching architecture for the new fabric architecture solidified, the next step was to evaluate cabinet options with the patented zero U capabilities using Siemon’s VersaPOD® cabinets. The study compared two rows of traditional 24-in. server cabinets flanked with wider end of row cabinets to support the new switching architecture to a full pod of VersaPOD® cabinets supporting both server and end of row switching environments.
Using a wider VersaPOD cabinet with zero U vertical space shared between two bayed cabinets (see Figure 1), PDUs are installed every other cabinet bay and whip/breaker or bus connections decrease by half compared to individual cabinet power connections. Copper and fiber are also shared between adjacent cabinets as shown below.
The savings realized compared to individual cabinet runs are as follows:
- Half the power whips are required as two adjacent cabinets share PDUs.
- Sharing PDUs between cabinets result in significantly fewer stranded ports and stranded power
- Copper and fiber ports are shared between two cabinets so individual copper and fiber panels and horizontal wire managers are not needed in each cabinet.
- Copper and fiber run vertically therefore copper and fiber jumpers are the shortest length resulting in roughly 40% savings in patch cords and jumpers over traditional panels mounted at the top of cabinets.
- Slightly fewer wider cabinets are deployed with higher power supporting the same number of servers and switches per pod.
A single row layout for each option is shown in Figure 1.
This section of the evaluation relates only to the passive costs associated with housing servers and switching environments and is comprised of cabinets, cabling, PDUs, and power whips. The total figures comparing the two options as shown in Figure 2 are as follows:
|
OPTION 1 |
OPTION 2 |
Total cabinets and cabling |
$ 462,908 |
$ 426,798 |
PDUs and whips |
$ 114,026 |
$ 62,605 |
Total cost per two- row pod |
$ 576,934 |
$ 489,403 |
Support cost/server |
$ 2,003 |
$ 1,716 |
Cost per 10 pods |
$ 5,769,347 |
$ 4,894,039 |
Total savings |
$ 875,308 |
The total costs of the cabling and cabinets are lower with VersaPOD, but the real savings is achieved in the power supporting the active equipment in the cabinets. With fewer power whips and PDUs the demonstrated savings is 54%. Across the 10 pods to be deployed in the second phase of the data center, the total savings on cabinets, cabling and power is $875,308. Assuming 288 servers per 2 row pod, the allocated cost per server decreases by 15% with VersaPOD. Not included by worth mentioning is the fact that as the PDUs support two cabinets, greater port utilization is achieved and the electrical losses throughout the upstream power distribution lines will be realized. Fewer power whips and breakers can mean fewer Remote Power Panels and large PDUs taking up valuable data center floor space.
In a similar exercise, comparing a high density model for phase II with 720 servers per pod the numbers are as follows:
|
OPTION 1 |
OPTION 2 |
Total cabinets and cabling |
$ 1,031,749 |
$ 769,133 |
PDUs and whips |
$ 252,161 |
$ 123,100 |
Total cost per two-row pod |
$ 1,283,910 |
$ 892,232 |
Support cost/server |
$ 1,783 |
$ 1,239 |
Cost per 10 pods |
$ 12,839,100 |
$ 8,922,324 |
Total savings |
$ 3,916,776 |
The higher density model shows an even lower supporting cost per server and a greater savings over the low density evaluation. Both evaluations completely dispel the notion that more cabinets with lower power per cabinet are less costly to deploy than wider cabinets with higher power density per cabinet. The aesthetics of properly dressed power and communications cabling are enhanced with larger cable management areas found in wider cabinets.
For the phase II deployment, the combined $9,727,411 savings identified by changing to a BASE-T architecture, and deploying wider cabinets result in over a million dollars in CAPEX savings in a low density scenario, and even greater in a high density configuration. Year over year OPEX savings will also be realized through lower power of the switching infrastructure, fewer power losses, less stranded power ports and less stranded power.