Raritan IT team recently decided to change our naming and association schemes in Power IQ. This change reflects the new level of power usage details IT wants to track, and how we plan to utilize such information to measure energy savings from the investment in replacing old servers with new, more energy efficient IT equipments.
In prior Power IQ naming scheme, we name the PX outlets by “phase-circuit-rack-L/R”. From this naming scheme, one can easily tell that it was a PX-centric thinking. Using Power IQ, IT was able to easily know the total data center power consumption, as well as power usage by circuit and by rack. The power consumption by individual IT devices was interesting but not as important for power capacity planning, especially when the servers were largely 1U/3U rack-mount servers.
When IT deployed Cisco UCS blade system, this focus has changed. The blade chassis is a device with high power density, IT has to track and monitor that carefully to ensure power capacity. And, the introduction of UCS blades was to start our transformation into Unified Computing, with the additional benefit of energy efficiency over the existing servers it will replace. IT would like to quantify the actual energy savings from this entire change as well as incremental savings per server decommission. As IT starts to migrate each old physical server as virtual servers onto the UCS blades, we would like to measure the delta between increased power consumption on the blade chassis and the reduced power consumption from decommissioning the old physical server.
Power IQ can help us do that easily if we change from PX-centric view to IT-device-centric view. With this change, IT can still obtain power usage per rack and per circuit from Power IQ; but it now becomes much easier to track power usage per device.
Between now and the end of this year, we will migrate all the old servers onto UCS and decommission them. Once we complete the entire migration, IT will know exactly the energy saving from this investment. For sure, IT wouldn’t introduce UCS just for energy efficiency if we didn’t have needs to increase computing capacity. But it will be very interesting to compare the actual energy saving from this investment with our original estimate based on server specs and vendor advices.
Hi Allen, would love to see your numbers some day. What Cisco’s competitors always fail to measure is the impact of their/our technology across the data center: ie. especially the impact on network gear.
If you scale UCS out to 112 blades then you will only need the two fabric interconnects verses ten times as many switches that competitors need. All those ports and switches suck up power, as we know, but nobody (apart from Cisco) seems to do that math.
Looking forward to seeing more,
Steve
I wrote on this blog last time that Raritan IT will monitor the incremental power consumption of the Cisco UCS we deployed, while we gradually migrate some rack-mount physical servers as VMs onto the UCS blades. The following results reflect our migration activities between 2010-08-05 and 2010-10-20. We have 2 UCS chassis that we call UCS1 and UCS2, and both of them are installed on our Cisco Rack 4. During the time period from 08/05 to 10/20, we had the following physical activities:
08/11: Turn on two blades on UCS for engineering testing
08/15: Turn on 2nd 10Gib port on Fabric Interconnect
08/30: Remove 2 VMware ESX Servers (2 Dell PowerEdge 1950)
09/01: Turn 1 additional blade on UCS2 to measure its effect on power consumption
09/03: Remove SharePoint Test Server (a Dell PowerEdge 2850) and DNS2 (a 1U PC-server)
09/14: Migrate BES as a VM onto UCS blade (an HP Proliant DL140)
09/20: Migrate Exchange Server (a Dell PE1950) as a VM onto UCS blade; turn testing blade off
10/11: Turn one UCS blade on for Oracle JDE remote DR site
From Power IQ data reading we have the following observations:
1. Each UCS blade we have consumes roughly 4.3 KWh per day. This is somewhat consistently demonstrated from measuring the total power consumption of the observed systems including the 2 UCS chassis and the non-UCS physical servers. On 08/11, the measurement was 334.6 KWh. On 08/12, 24-hours later after turning on 2 UCS blades, the measurement was 339.7KWh. Since then the UCS power consumption stayed at that level until the next even on 08/15. This 4.3KWh incremental power consumption per UCS blade per day was also consistently measured between 09/01 and 09/02 when IT turned on the 3rd blade on UCS2 to measure incremental power consumption.
2. Migrating mildly loaded physical rack-mount servers as VMs onto the UCS blades typically doesn’t increase much power consumption. We Installed VMware ESX host on UCS and migrated two physical application servers onto UCS blade, and we couldn’t even find material difference before and after adding these 2 VMs onto UCS. But the power savings from decommissioning physical servers can be clearly seen. On 08/30, we removed 2 ESX host physical servers (Dell PowerEdge 1950), the total power consumption dropped by 6.55KWh a day. Then on 09/03, we removed another PowerEdge 2850 (a Sharepoint test Server) and a 1-U PC server (DNS); and we saw a drop of power consumption by 7.2KWh a day. Therefore, we see a clear power saving advantage from migrating physical servers into VMs.
3. The new version of Power IQ is very helpful in data exporting and giving power consumption insights at various levels of details. We captured several diagrams from Power IQ to see the trend over time, then right from there we export the data into Excel spreadsheet, and we can examine the day by day details in there.
Long time reader / first time poster. Really enjoy reading the blog, keep up the good work. Will most definitely start posting more in the future.