BladeSystem - General
cancel
Showing results for 
Search instead for 
Did you mean: 

Running power consumption

redladmick
Occasional Visitor

Running power consumption

Can any please tell me what the running power consumption is on a full 16 bladed rack? c7000 enclosure I have secured space in a data centre and they tell me i am limited to 3.5kv per rack, surely a fully bladed rack consumes more than that?? Any help would be appreciated. Thanks Mick [Updated on 6/27/2008 6:49 AM]
4 REPLIES
James Henry_5
Occasional Visitor

Running power consumption

Hi Mick, You can accurately get the power requirements for any blade solution using the BladeSystem sizer tool available under the get Started section on www.hp.com/go/bladesystem The direct link is http://h71019.www7.hp.com/ActiveAnswers/cache/347628-0-0-0-121.html The truth is that it could be anywhere between 2.5kW and 4 kW dependingon the spec you choose. The good news is that with HP BladeSystem you also have the capability to cap the power on the enclosure and even on a per blade basis should you wish, so you never exceed your limits.

Running power consumption

Make sure they are providing 208V, ie, make sure this is not 2 X 120V circuits, etc. The previous poster is correct if you are using low watt CPU and small configs. On average though for over 4 DIMM's and non low watt CPU's a fully loaded enclosure with HH is closer to 3.5 - 5.5Kw. I have seen some that can peak 6Kw. True you can power limit this in OA but if you limit to 3.5 on a 4.5 or 5Kw config you are most likely going to lose the ability to power on a large number of blades. 3.5Kw is not even 2 Power Supplies (2250W each at 208V). Just some thoughts. If I were you I would try to get more power. Bare minimum 30A / 208V circuit with 24A/208V PDU giving you 4992W usable for the enclosure. Shawn
chopper3
Frequent Advisor

Running power consumption

As the previous posters say enclosure power can be managed by matching processors, processor model, memory amount and type, disks and mezz/interconnects to what is required by your applications. That said we have lots of full enclosures (with 16 SQL servers using dual E5440's, 12GB, 2 x 146, Emulex ) and they generally draw about your power budget, 3.5Kw - if you don't need such relatively high-spec blades then you'll come in under that. Two things to state however. Firstly HP's Blade Sizer is famously pessimistic when it comes to power consumption planning, I'm not too bothered, I'd have written the software that way too if I were them but I would always use a utilisation figure of around 55-60% to see a realistic view of power use. Of course if you plan to keep them 100% utilised by some form of cruel looping code then your mileage will be different. Secondly, power capping works, but has a problem - and it's one I've had a HP power guy go red in the face with embarrassment about. If you set your enclosure power cap to 3.5Kw it won't go over that, which is good - but servers need more power on startup than once they're stable - if you try to bring up all, or many, of your servers at once or close together and their total stable power draw is close to your capped limit then some servers won't be allowed to start and will raise an alert. As an example we have lots of 10/11 server C7K enclosures (single E5345's, 10Gb, 2 x 146GB, nc373 mezz) and when stable they hover at about 2.9/3Kw - with a 3.5Kw cap two or three servers wouldn't come up due to lack of available power. We had to set the cap to 5.4Kw to reliably bring the lot up at once. Hopefully your 3.5Kw limit is a cooling one, not a power-draw one, in which case don't worry about it, just set the cap to 5-6Kw and keep an eye on the stable draw - if more than 3.5Kw will 'pop' your PDU then you've got a problem. As I say, HP know of this problem and I'm sure that a future OA code update will allow you to define a 'bounce-order' but until then this is a very real problem. That said blades are great :) Regards, Phil. [Updated on 8/9/2008 1:42 PM]
Tony Harvey_1
Occasional Advisor

Running power consumption

The problem with the power limit function is not really a problem it is functioning as designed. The objective with that feature was to ensure that no matter what if you set the power limit the enclosure would not exceed that limit So if you for example set the limit to 3.6kW which is about what a 16A breaker can support, the enclosure would not exceed that limit and pop the breaker even when starting up all 16 blades. This of course has the nasty side effect as described that in some cases not all blades will power up, but we didn't pop the breaker. What the power limit is doing is interacting with the way we allocate power in the enclosure. When the enclosure comes up the OA looks at what Power Supplies, Interconnects and fans are installed in the enclosure. It then looks at the redundancy that has been set and from this works out how much power is left over for each blade. An example would be with 6 PSUs and N+N redudnancy I have 6750W (3 x 2250) as starting power pool, from which interconnects and fans need to be subtracted whatever is left is available for blades. Setting the power limit reduces that initial pool so if you set a power limit of 3.5kW you are reducing that initial pool. This interacts with the way the blades are allowed to power up as well. Before each blade is allowed to power up it sends a request for power to the OA, this must be granted by the OA otherwise the blade will not power up. Initially each blade requests a max power number from the pool. This number is burned into the blade firmware and represents a worst case config. If available this Max Power number is allocated out to the blade and the OA subtracts it from the available pool of power. The OA then goes to the next blade power-on request. While the OA is letting other blades power on, the intial blade is going through POST, as part of this it does a "burn test" that figures out what's the actual power consumption of this blade as it's configured rather than the max number. This new number is sent to the OA so it can readjust the available pool of power. We try try to get as many blades powered up as possible within the power limit, but the objective is to not allow something to happen that would cause the breaker to trip. The OA will prevent blades from powering up to ensure this. We have enhanced some features in the OA recently (current version is 2.25) where we provide the Device Power Sequence capability which allows you to set time delays on blades so you can guarantee which blades will come up first