HP c-Class blades take energy efficiency to a new level
HP's newest c-Class blade chassis 16-blade system has brought energy efficiency design to a whole new level in server infrastructure. The unique design allows the system to shut down up to 5 of the 6 2250 watt power supplies to maintain a peak load on the remaining power supply which translates to more than 90% power supply efficiency. If power supply redundancy is needed and configured on the chassis, it will leave 2 of the power supplies on in case one of them should fail. This would translate to less than 140 watts of average power consumption for each of the 16 2-socket Intel or AMD based servers while they are operating at lower utilization levels.
Normal servers with independent dumb power supplies don't even come close to this efficiency level and just throw wasted heat in to the data center which jacks up the cooling requirements. Rack mount servers have dumb oversized power supplies which can't pool a single power supply over a DC (Direct Current) distribution system across 16 servers. To keep the power supplies from over loading, the servers will inform the chassis of their work loads so if there is a need for more computational power that exceeds the power supply capacity, then more power supplies will come on line. On the other hand, power draw limits can be placed on servers so that they don't exceed a maximum power draw. Even though this may prevent servers from reaching their full operating potential any time they want, it rarely affects application performance since CPUs rarely maintain peak loads and the power supplies can always come on line quickly to allow for full throttle. Even when a single server does peak out in load, it doesn't mean the other servers will so the over all power consumption may not go up that much across 16 servers. All of this is managed through a web interface for the chassis system and I was relieved to find that no Java Runtime Engine was required for the management package.
One other interesting energy saving aspect of the c-Class blade system is the cooling system. 10 miniature jet engines fans are used in place of many more smaller fans which results in even more energy savings. A 1 U server for example might use 4 or more fans and if we multiply that by 16 servers, we're talking 64 fans versus 10 powerful miniature jets. The 10 jet fans are also easily hot-swappable from the back.
The rack density of this blade server system also doubles compared to 1U servers. To easy the migration from 1 or 2 U rack-mount servers, the blades are image compatible which means you can deploy the same server images and you won't need to recreate or recertify a whole new hardware standard. The skill sets and processes that have already been developed can be directly translated to the blade servers. The amount of rack mounting and cabling work is reduced by 16 times which puts a smile the face of anyone who has ever spent long days and nights rack mounting servers. From a hardware cost standpoint, the pricing of blade servers are now competitive with stand alone rack mount servers and sometimes even better. Only generic white box servers are cheaper at this point but the ease of management and energy efficiency should cause any data center manager to think long and hard about blade servers.