Scaling Energy Adaptive Applications for Sustainable Profitability - PowerPoint PPT Presentation

Scaling Energy Adaptive Applications for Sustainable Profitability Fabien Hermenier , Giovanni Giuliani, Andre Milani, Sophie Demassey

Let existing and new data centres become energy adaptive Adapting the power consumption to the availability of renewable energy Being adapted to the requests of a Smart City Energy Management Authority

How to reconcile competing objectives ? An economical approach to avoid overcommitment Energy authority Client regulate best energy usage performance

[ucc 2015] energy adaptive applications forecasts web service Easc energy application descriptor Carver ... providers elected working modes video transcoder Easc sustainable objective smart city authority Orchestrates EASC for profitable sustainability Optimises over 24-hrs time window (96 time-slots of 15 minutes) Called every 15 minutes to accommodate uncertainty

The energy providers Per time-slot data or forecasts for the next 24 hours % issues from renewable sources price capacity

Smart city energy authority Regulate energy usage through contracts day-to-day periods At least X% renewable, Objective Power budget, … Incentive Penalty function (flat, linear, composite)

EASC characterisation

Service Level Objectives daily-based deferrable workloads slot-based non-deferrable workloads flat or linear pricing policies

The working modes discretise elastic applications states performance level power consumption transition cost actuator manual or automatic calibration

3 use cases inside DC4Cities Webservice, video transcoding, e-health

The underlying problem For every time-slot find 1 working mode per EASC using available energy (viable dispatch) 96 NP-hard bin-packing problems to solve + transition costs + various Energy Authority policies + Instant and cumulative objectives + generic pricing policies = I dont want a heuristic full of corner cases

An Open-Source java library for constraint programming deterministic composition high-level constraints the right model for the right problem

Modelling - TLDR; Working mode -3 -5 -3 12 11 10 time 9 8 7 6 5 An automaton to model each EASC life-cycle Counters to accumulate daily incomes Energy to use is dispatched among the sources Energy authority pricing policy over the cumulative daily usage ∞

Reconciliations “ Trading energy, performance and energy authority conformance to maximise the daily running costs

Evaluating Carver real deployments 3 testbeds running production softwares HP Lab @ Milan CSUC @ Barcelona Trento @ Italy Web service Mixed EASCs e-health Video transcoding Openstack Bare metal openNebula stack

Testbed 20 HP moonshot cartridges in 2 chassis no virtualisation layer 20 Watts peak The grid renewable power (%) 100 un-pure energy 75 market price 50 25 0 16/01 17/01 18/01 19/01 20/01 Date The green 8m2 of solar panels 600 Power (W) 4 typical days 400 from a 1 y. data collect 200 0 16/01 17/01 18/01 19/01 20/01 Date Energy authority expectations 65% renewable target 100 € / pp.

e-learning courses for Entrepreneurs 3000 production software stack 2500 with 3 EASCs Request/s 2000 1500 Injectors mimic the 1000 production workload 23:00 05:00 11:00 17:00 23:00 Time E-learning, G-learning Web service cumulative SLOs instant SLO 3 working modes 6 working modes penalty from a linear function penalty from a step function

Carver Perf Sustainable profitability Production mode VS Cache built at midnight Lowest suitable working mode Green for the WebSite Most e ffi cient working mode when renewable energy is here Lowest otherwise

Perf 500 application 400 Ignores energy availability Watts E − learning 300 G − indexing 200 Website 100 Workload a ffi nity: 0 human work during days 16/01 17/01 18/01 19/01 20/01 Date application application 400 E − learning 400 E − learning Watts Watts G − indexing G − indexing 200 Website 200 Website 0 0 20/01 19/01 18/01 17/01 20/01 16/01 19/01 18/01 17/01 16/01 Date Date Green Carver s u n h e w t o l l o e s f v i t i c t i c h a B a t r y b i n a i s e e n G r

Carver do not over-commit 17/01/15 18/01/15 19/01/15 20/01/15 70 max achievement 67.7 65.6 64.2 Carver sticked to renewable % 60 58.5 57.9 57.3 56.7 the green threshold 55.6 53.9 50 47.4 46.8 45.7 40 f r n f r n f r n f r n r r r r e e e e e e e e e e e e v v v v p e p e p e p e r r r r r r r r a a a a g g g g C C C C % renewable 17/01/15 18/01/15 19/01/15 20/01/15 1999 1999 1999 1999 1999 1955 1933 1936 2000 SLO penalty (euros) Carver sticked to 1466 1500 1352 1305 the SLO 1072 1000 500 0 perf green Carver perf green Carver perf green Carver perf green Carver Customer incomes

17/01/15 18/01/15 19/01/15 20/01/15 Running cost (euros) 2000 expense energy SLO 1000 SMA 0 f n r f n r f n r f n r r r r r e e e e e e e e e e e e v v v v e e e e p p p p r r r r r r r r a a a a g g g g C C C C Green neglects the clients Perf neglects the energy authority Carver trades

17/01/15 18/01/15 19/01/15 20/01/15 Running cost (euros) 2000 expense energy SLO 1000 SMA 0 f n r f n r f n r f n r r r r r e e e e e e e e e e e e v v v v e e e e p p p p r r r r r r r r a a a a g g g g C C C C The balance is still slightly pure performance oriented nothing to do to please the energy authority @Day 1 (natural workload a ffi nity) @Day 4 Only minor trading possibilities

17/01/15 18/01/15 19/01/15 20/01/15 Running cost (euros) 2000 expense energy SLO 1000 SMA 0 f n r f n r f n r f n r r r r r e e e e e e e e e e e e v v v v e e e e p p p p r r r r r r r r a a a a g g g g C C C C Economically speaking Energy price does not play a role every price <<< pricing policies to avoid bankruptcy small data centres human resources / software support dominates

Lesson learned CP composability helps but data makes the problem being generic is costly static analysis for stronger models require favorable variable working modes energy-e ffi cient hardware hardware and software a multifacet tool PV array sizer cost modelling prospective deployment

Sustainable profitability to motivate energy transition Carver is looking for sun and profit A flexible solving algorithm to cope with the problem variability Conciliation possibilities validated on industrial testbeds A multi-faceted tool

http:/ /www.dc4cities.eu deliverables, scientific publications, trial results, software repositories