2. Agenda
BrainServe in very brief
The high-density choice: back in 2008
Cooling technologies: low-density and high-density
Water-based cooling
BrainServe example
High-density and energy efficiency
Q&A
Swiss Big Data User Group Meeting – 24.06.2013
3. BrainServe in very brief
Swiss based independent datacenter in the northwestern part of Lausanne
Physical housing space for IT equipment
Collaboration with IT providers and telecom operators
Dedicated plot owned by BrainServe
Location with a low risk profile
Datacenter design and building « from scratch » (2008 – 2010)
Highly secure and resilient environment (2’000 m2 of housing areas)
1’000 m2 of «classical» air cooling
1’000 m2 of high-density liquid cooling
Swiss Big Data User Group Meeting – 24.06.2013
4. The high-density choice: back in year 2008
Emergence of data-centric universe
Already more data, and still increasing
Storage, processing needs and access requirements are increasing accordingly
Hardware evolution: power densification
Historically (1970 – 2005): smaller transistors used less electricity to operate
exponentially better performance & density for a constant power envelope (“free energy”)
Since 2005: continue to make transistors smaller, but they use similar electricity to operate
chip energy consumption is shooting up (end of “free energy”)
Electrical input = heat output
increase in computing power = increase in energy consumption = increase in heat dissipation
It is often not possible nor economical to spread the load across the space
increase in computing power = increase in heat density requires high-density cooling
Swiss Big Data User Group Meeting – 24.06.2013
5. Cooling technologies: low-density
Swiss Big Data User Group Meeting – 24.06.2013
Traditionally: air is the primary heat transport media
Heat exchanger are away from the IT racks
Air is blown into the plenum under the raised
floor and comes out in front of the racks
(preferably on a cold aisle containment)
In standard configuration (traditional
room height): 4 kW/rack on average
Technical and operational constraints
may lead to very disparate densities and
hot spots that are difficult to handle
6. Cooling technologies: high-density
High-density applications require typically
10 kW to 20 kW per rack
Different manufacturers and designs for high-
density cooling: in-row cooling, in-rack cooling,
overhead cooling, rear door cooling
Identical central feature: bring the heat
exchanger closer to the server rack
Reduced volume of air to be moved
Shorter and more predictable air flows
Handling of hot spots easier
Better use of the cooling capacities
Most of the products require connections
to a chilled water loop
Swiss Big Data User Group Meeting – 24.06.2013
7. Water-based cooling
Why do we use water?
Air is abundant and easy to handle, but not a very good heat conductor
To transfer more heat we need to dramatically increase the volume of air
transported
Water is a much better heat conductor
Swiss Big Data User Group Meeting – 24.06.2013
Source:AndréOppermann,SWINOG-26
8. BrainServe example (1)
1’000 m2 (50% of IT space) equipped for high-density
N+N redundancy (production, distribution and cooling units)
Necessary to cope with rise of temperature in case of failure
In-row cooling with hot aisle containment
Designed for 10 kW/rack on average,
easily scalable to 50 kW/rack
Swiss Big Data User Group Meeting – 24.06.2013
9. BrainServe example (2)
Great modularity
Positioning of the rack cooling matches the specific needs
Proportions of IT racks and cooling units are tightly coupled
Automatic chilled water network isolation
in case of major failure
Water leak detection system
10 MVA electrical feed
Power distribution with 400 A and 630 A busbars
Fire detection and automatic extinction
BrainServe is an industrial partner for the EXTREME
(Energy- and thermal-aware design of many-core
heterogeneous datacenters) research project with
the EPFL that look for future energy efficient design
for IT and datacenters
Swiss Big Data User Group Meeting – 24.06.2013
10. High-density and energy efficiency
Energy efficiency is a major concern for datacenters
Among others, high-density is a factor of efficiency improvement
A datacenter is most efficient when it is fully loaded, and the IT load has the
biggest impact on the global efficiency (bigger than e.g. the outside temperature)
Densification of the IT load improves the energy efficiency: the more energy we use, the more
energy we save
By bringing the heat exchanger closer to the heat source, we keep the air volume
to be moved as low as possible
Power consumption is proportional to the cube of the fan speed
By increasing the ΔT on the heat exchanger (e.g. with hot aisle containment), we
lower the air volume to be moved for a given load
High-density does not mean lower temperature in the IT room
ASHRAE recommended environmental envelope is independent from the power density
High-density means more re-usable heat
Swiss Big Data User Group Meeting – 24.06.2013
11. Gabriel Boissonnard
T +41 21 637 69 04
M g.boissonnard@brainserve.ch
W www.brainserve.ch
THANK YOU FOR YOUR ATTENTION!
Swiss Big Data User Group Meeting – 24.06.2013