All the successful managed server and data center space providers out there (like Berkeley IT) are suffering a shared problem. A shortage of density and capacity; those who havent really kept up with the market see power delivery (wattage not CPU) and cooling as the major issues with the data center space, well I dont agree, I do recognise that it is an issue, but with some of the new product releases due this year, this problem should be one of the past in some cases. For the purposes of this discussion I am going to commit some physics fopars, I realise its not this simple, TRUST me, I realise it isnt this simple… i’ve speant over a year working on different data center build plans!
Essentially the more power (wattage) a system draws, the more heat it will produce, and thus, as a data center operator the more you have to cool. The issue is that air has an effective cooling density, there are arguements over what this is, but 300w/sqft can be cooled quite effectively, both in terms of additional power required to drive large HVAC unit, as well as cost of cooling, and as everyones going green crazy, lets not forget the environmental impact. Well this market state was brought about for 3 reasons;
- Demand – people wanting more and more powerful systems (blade centers etc)
- Hardware companies – bringing inefficient servers to market
- Stupid people – thinking that the way to get more CPU power is by putting in uber blade centers
Hardware vendors have now caught up with the market going “oi, we dont want you 3amp drawing boxes” *cough* Apple XServe *cough*, why would you put in a box that draws 3amps when a 1.5amp draw server from HP can do the same job? You see its an escalating problem, servers use more power, data centers charge for more power, then have to provide more cooling capacity, which in its own right draws power. Also typically big blade centers require 32amp power feeds, the cost to provision this in most data centers is `quite large`, and getting that power in London is almost out of the question.
The market has already shifted back to days of old, where you had entire floors dedicated to processing, think huge VAX300 systems and multinode IBM mainframes. With utility computing the same processing power can be achieved without the large power draw, cooling requirement, and environmental impact that we’ve seen in recent years. The emergence of virtualisation is another critical step in this arena, with the ability to start `servers` on demand, based on system loads, we no longer need to have redundant physical capacity sitting there when its not required.
Berkeley IT is effectively delivering 10.5Ghz of processing power on a nominal 220w draw (thats under an amp – remember American readers, UK 240V != 120V). How are we pulling this off, here are some of the systems that we are looking to deploy for our larger demanding customers. Why am I sharing this? Well its old news as far as I’m concerend, Berkeley IT is currently developing newer, denser and more efficient models, staying ahead of the curve and all that;
The Octimod MPC7-1500 – think blade center without the power draw 
Image courtesy of mini-itx.com
7 processing modules, each is hot-swappable, and contains a 1.5GHz C7 powered VIA EPIA EN 15000 motherboard. The modules boot over the network from a master node (or from multiple master nodes). SATA RAID modules provide shared storage, with a rack mounted distributed DC power supply providing power for each module. 8 of these 4U Octimod units can be mounted in a 42U rack, and as they are only 300mm deep, we can mount them back to back. So thats 16 units, each delivering 10.5Ghz of processing power at 220W full loading. So a per rack draw of 14.6 amps (excluding the switch). Berkeley IT can provision 16 amp racks for roughly £800 a month, compare that to what you are currently spending on your processing power and blade centers! IE: Stick that in your pipe and smoke it, for as always, its Berkeley IT for the win
LinITX.com’s Quad CV700A 1GHz
Image courtesy of LinITX.com
Essentially this if a 1U rackmount case with 4 independant systems. Each has 3 onboard network adapters… why I’m not quite sure, we rarely make use of two, but hey! They run at 1Ghz and can take up to 1Gb of RAM. Each machine has its own independant 2.5″ hard disk. Oh and they are under a grand each!
Polymath. Serial entrepreneur turned VC, now sitting on both sides of the table, talking tech, finance, and motorbikes.