5/31/14
Autonomic Clouds Part II
Omer Rana (Cardiff University, UK) Manish Parashar (Rutgers University, USA)
Defining the landscape • Analysis of exis ng published work – Give poten al research direc ons • Various uses of autonomics – will focus on: – Auto scaling and elas city – Streaming analy cs – Integra ng autonomics into applica ons • Integra ng autonomics with – Exis ng Cloud middleware – Distributed Cloud landscape (e.g. GENI Cloud)
1 5/31/14
Cross layer Intelligence
Cross layer Intelligence • Applica ons can exhibit dynamic and heterogeneous workloads – Hard to pre-determine resource requirements • QoS requirements can differ across mul - tenancy applica ons – Batch vs. real me, throughput vs. response me • Integra ng local resources with Cloud provisioned resources – Cloud “burs ng” (when and for how long) – Data sharing dynamically between the two – Cost implica ons for long term use
2 5/31/14
From Chenyang Lu (Washington Univ. in St Louis)
Links with Control theory • Applying input to cause system variables to conform to desired values – o en a “set” or “reference” point – E-commerce server: Resource alloca on? à T_response=5 sec – Embedded networks: Flow rate? à Delay = 1 sec – Power usage: Energy? à Consump on < 250Wa s
• Provide QoS and related guarantees in open, unpredictable environments
• Various modelling approaches: – Queuing theory (very popular) – no feedback generally in queuing models; hard to characterise transient behaviour overloads – Other approaches: Petri nets and Process algebras – O en a design/tune/test cycle – repeated mul ple mes
From Chenyang Lu (Washington Univ. in St Louis)
Open-loop control
• Compute control input without con nuous variable measurement – Simple – Need to know EVERYTHING ACCURATELY to work right • E-commerce server: Workload (request arrival rate? resource consump on?); system (service me? failures?) • Open-loop control fails when – We don’t know everything – We make errors in es ma on/modeling – Things change
3 5/31/14
From Chenyang Lu (Washington Univ. in St Louis)
Feedback (close-loop) Control
Controlled System Controller
control control manipulated input Actuator function variable
error sample controlled + Monitor - variable
reference
From Chenyang Lu (Washington Univ. in St Louis)
Feedback (close-loop) Control • Measure variables and use it to compute control input – More complicated (so we need control theory) – Con nuously measure & correct • Ecommerce server: measure response me & admission control • Embedded network: measure collision & change backoff window • Feedback control theory makes it possible to control well even if: – We don’t know everything – We make errors in es ma on/modeling – Things change
4 5/31/14
From Chenyang Lu (Washington Univ. in St Louis)
Control design methodology
Modeling Controller analytical Dynamic model Control algorithm Design system IDs
Satisfy
Requirement Performance Specifications Analysis
From Chenyang Lu (Washington Univ. in St Louis)
System Models
• Linear vs. non-linear (differen al eqns) • Determinis c vs. Stochas c • Time-invariant vs. Time-varying – Are coefficients func ons of me? • Con nuous- me vs. Discrete- me • System ID vs. First Principle
• System Goals: – Regula on (e.g. target service levels) – Tracking (measuring devia on from a target, e.g. change #VMs) – Op misa on (e.g. minimize response me)
5 5/31/14
VM consolida on • A commonly referenced problem in Cloud compu ng – Server cost the largest contributor to overall opera onal cost • Data centers operate at very low u liza on – Microso : over 34% servers at less than 5% u liza on (daily average). US average 4%. • VM Consolida on increases u liza on, decreases idling costs • However VM consolida on can cause interference in the memory hierarchy (e.g. due to sharing of cache between cores or memory bandwidth)
VM Consolida on: PACMan • How much will each VM degrade when placed with other VMs? • Which and how many VMs can be placed on a server whilst s ll maintaining performance? • PACMan (Performance Aware Consolida on Manager) – Minimise resource cost (energy usage or #servers) – Use of an approximate (computa onally efficient) algorithm • Differen ate between: – Performance vs. resource efficiency (e.g. batch mode) – Eco mode: fill up server cores (minimise worst case degrada on e.g. MapReduce – minimise me of worst case Map task)
6 5/31/14
Parameters considered: “n” VMs, on “m” machines with “k” cores Consider a “degrada on” parameter as new VMs are added to a machine – iden fy an op misa on goal – e.g.
Alan Roytman, Aman Kansal, Jie Liu and Suman Nath, “PACMan: Performance Aware Virtual Machine Consolida on”, Proceedings of ICAC 2013, San Jose, USA (USENIX/ACM) PACMan – Approach • Start from an arbitrary ini al schedule • For all ways of swapping VMs, go to the schedule with smallest sum of maximum degrada ons • Limit total number of swaps to achieve convergence
7 5/31/14
Elas city • One of the key “selling points” of Cloud systems • Various approaches possible: – O en historical informa on useful (response mes, queue lengths, arrival rates and request demands) – Long-term, medium-term and short-term planning – VM alloca on and placement • Reac ve vs. proac ve approaches
Dynamic VM alloca on • Understanding “Elas city” the degree to which a system is able to adapt to workload changes by provisioning and de- provisioning resources in an autonomic manner, such that at each point in me the available resources match the current demand as closely as possible. • Can elas c provisioning capability be measured
“Elas city in Cloud Compu ng: What It Is, and What It Is Not” Nikolas Herbst, Samuel Kounev, Ralf Reussner, ICAC 2013 (USENIX)
8 5/31/14
Dynamic VM alloca on
• Scale up speed: switch from an underprovisioned state to an op mal or overprovisioned state. – Can we consider “temporal” aspects of how scaling up takes place • Devia on from actual to required resource demand – Measure devia on to influence the overall process • Role of predic ve alloca on for “known” events – i.e. know in advance how many VMs to allocate
Scaling Influences & Strategies
• Reac ve – Observed degrada on in performance over a par cular me window • Trace-driven – Based on a short-term predic on – Could make use of a “cyclic” workload pa ern • Model-driven – Use of a queuing/Petri net/dynamic systems model – O en parameters “observed” and tuned off line
9 5/31/14
An Example
“Elas city in Cloud Compu ng: What It Is, and What It Is Not” Nikolas Herbst, Samuel Kounev, Ralf Reussner, ICAC 2013 (USENIX)
Comparing alloca on
“Elas city in Cloud Compu ng: What It Is, and What It Is Not” Nikolas Herbst, Samuel Kounev, Ralf Reussner, ICAC 2013 (USENIX)
10 5/31/14
Elas city Metrics
“Elas city in Cloud Compu ng: What It Is, and What It Is Not” Nikolas Herbst, Samuel Kounev, Ralf Reussner, ICAC 2013 (USENIX)
Elas city Metrics … 2
“Elas city in Cloud Compu ng: What It Is, and What It Is Not” Nikolas Herbst, Samuel Kounev, Ralf Reussner, ICAC 2013 (USENIX)
11 5/31/14
H. Nguyen, Z. Shen, X. Gu, S. Subbiah, J. Wilkes, “AGILE: Elas c distributed resource scaling for Infrastructure-as-a-Service”, Proceedings of ICAC 2013, San Jose, USA (USENIX/ACM)
AGILE
AGILE
• Medium term predic ons using Wavelets • Use of an “adap ve” copy rate – Pre-copy live VM based on predic on – Avoids performance penalty – Does not requiring storing and maintaining VM snapshots – Can be undertaken incrementally – therefore avoids “bursts” in traffic when submi ng an en re VM (e.g. compared to “cold cloning” • Supports post-cloning auto-configura on
12 5/31/14
Suppor ng Elas c Behaviour • Variety of approaches possible: • Modelling decisions as a Markov Decision Process (TIRAMOLA successfully resizes a NoSQL cluster in a fully automated manner) • Use of classifier ensemble • Machine learning strategies (e.g. use of neural networks) • Rule-based (trigger-driven) approaches
“Automated, Elas c Resource Provisioning for NoSQL Clusters Using TIRAMOLA” Dimitrios Tsoumakos, Ioannis Konstan nou, Chris na Boumpouka, Spyros Sioutas, Nectarios Koziris, CCGrid 2013, Del , The Netherlands
Hybrid Approaches
• Use of different techniques for scaling up vs. scaling down – Reac ve rules for scaling up, regression-based techniques for scaling down – Reac ve rule: queue length of wai ng requests (but could be other criteria) – Predic ve assessment (use of queuing models) to dynamically trigger new VMs
13 5/31/14
TIRAMOLA
TIRAMOLA • Decision Making – cluster resize ac on according to the applied load, cluster and user-perceived performance and op miza on policy – Modelled as a Markov Decision Process (look for best ac on w.r.t. current system state) – User goals defined through a reward func on (mapping of op misa on goals) • Monitoring via Ganglia – Server + user metrics (via gmetric spoofing) • Cloud Management – Via euca2ools (Amazon EC2 compliant REST library) • Cluster Coordina on – Via remote execu on of shell scripts
14 5/31/14
TIRAMOLA • Formulates resize decisions as a MDP – State defined as #VMs, CPU usage, memory – Ac ons: add, remove or do-nothing (no-op) – Ac ons limited by a quan fier, i.e. add_2, add_4 (restric ons on these quan fiers) – Transi on prob. – based on if state is permissible or not (e.g. can exact number of VMs be added) – can be generalised to par al addi ons – Reward func on – r(s): “good ness” of being in state (s); r(s) = f(gains, costs) • MDP enables: – No knowledge of dynamics of environment is assumed – Learn in real me (from experience) and con nuously during the life me of the system
TIRAMOLA • Use of Q-learning (a type of reinforcement learning)
• Base calcula on of r(s) on a par cular arrival rate (of requests) and certain number of VMs • Collect results into a table – and use historical data to iden fy ac on and s’ (given s) • r(s) = f(latency, VMs)
15 5/31/14
AutoFlex • Use of monitoring to collect: – CPU, memory, network bandwidth, opera ng system queues, etc. • Controller (feedback mechanism) – Compares target with actual – Launches or terminates VMs • Controller is both reac ve and proac ve – Layer controllers that run periodically (short term planning) – Reac ve behaviour through ac ons for different resource types – Predictors a empt to es mate future u liza on – Mul ple predictors – with the use of a selector to choose
“Autoflex: Service Agnos c Auto-scaling Framework for IaaS Deployment Models” Fabio Morais, Francisco Brasileiro, Raquel Lopes, Ricardo Araujo, Wade Sa erfield, Leandro Rosa IEEE/ACM CCGrid 2013, Del , The Netherlands
16 5/31/14
AutoFlex … predictors • Keep CPU U liza on < 70% • Predictors used: – auto-correla on (AC), – linear regression (LR), – auto-regression (AR), – auto-regression with integrated moving average (ARIMA), and – the previous u liza on measured (dubbed Last Window, or simply LW) – Ensemble using all of the above • Metrics: – Hard viola ons: capacity not enough to handle demand – Cost: auto scaling vs. over provisioning (knows highest demand and sta cally allocates resources)
AutoFlex … predictors
• Based on 265 traces from HP users
17 5/31/14
YinzCam (CMU)
YinzCam is a cloud-hosted service that provides sports fans with • real- me scores, news, photos, sta s cs, live radio, streaming video, etc., • on their mobile devices • replays from different camera angles inside spor ng venues. • YinzCam’s infrastructure is hosted on AmazonWeb Services (AWS) and supports over 7 million downloads of the official mobile apps of 40+ professional sports teams and venues within the United States.
h ps://www.cmu.edu/homepage/beyond/2008/spring/yinz-cam.shtml
YinzCam – demand profile
week-long workload for a hockey-team’s mobile app, illustra ng modality and spikiness. The workload exhibits the spikes due to game-day traffic during the three games in the week of April 15, 2012
18 5/31/14
Auto Scaling strategies • YinzCam provides an example of various streaming applica on requirements • Some events are predictable: – Poten al workload during a game (historical data) – “in-game” vs. “non-game” mode – Some events are not (e.g. likely demand during a par cular gaming event) • Other scenarios: – Unpredictable scale up (e.g. observed phenomenon trigger in a sensor network) • Generally: over provision during game event
Scale up/down policies
• CPU usage threshold à trigger new VM – YinzCam (30% CPU usage over 1 minute) • Aggressive scale up, cau ous scale down – Overcome VM alloca on overheads – Poten al for oscilla on in the system (at next CPU check) • Example policies: – Mul plica ve Increase, Linear Decrease – Linear Increase, Mul plica ve Decrease • Inspira on from TCP and other conges on control mechanisms
19 5/31/14
Baseline is “manual” scaling
“To Auto Scale or not to Auto Scale”, N. D. Mickulicz, Priya Narasimhan, Rajeev Ghandi, ICAC 2013 (USENIX), San Jose, CA
Dynamic SLAs • Applica ons on mul -tenancy infrastructure – With changing applica on demands (e.g. must respond to unpredictable events) • Prevent “over specifica on” of service level demands – User might make an ini al assessment of likely demand (“first stab” at likely app. behaviour) • Provide SLAs that are “machine generated” – Based on predic ve usage between applica on classes – Offers made to users based on “likely” demand profile – May u lise resource thro ling strategies (cgroups in Linux – control groups that limit resource consump on)
20 5/31/14
Dynamic SLAs in OpenStack
“A Research and Development Plan for Dynamic Service Level Agreements in OpenStack”, Craig Lee, ITAAC workshop alongside IEEE/ACM U lity & Cloud Compu ng Conf, Dresden, December 2013
Admission Control • Reaching QoS of applica ons is o en strongly driven by admission control strategies • Admission control in large-scale cloud data centres influenced by: – Heterogeneity à performance/efficiency – Interference à performance loss from high interference – High arrival rates à system can become oversubscribed • Paragon and ARQ could be two approaches – Paragon: heterogenity and interference aware scheduler • ARQ: Admission control strategy. – Use of Paragon to classify applica ons into mul ple request queues – Improve u lisa on across mul ple QoS profiles “ARQ: A Mul -Class Admission Control Protocol for Heterogeneous Datacenters”, Chris na Delimitrou, Nick Bambos and Christos Kozyrakis h ps://www.stanford.edu/group/mast/cgi-bin/drupal/system/files/2013.extended.arq_.pdf
21 5/31/14
Paragon (Stanford)
Sources of Interference (SoI) benchmarking • Targeted microbenchmarks of tunable intensity that create conten on in specific shared resources • Introduce conten on in: processor, cache hierarchy (L1/L2/L3 & TLBs), memory (bandwidth and capacity), storage • Run applica on concurrently with microbenchmark – Progressive tune up intensity un l QoS viola on – Associate a “sensi vity score” with applica on (i.e. sensi vity to interference) • Similarly, Sensi vity to running applica on – Impact of running applica on on micro-benchmark – Tuning up applica on intensity un l 5% degrada on on benchmark (compared to execu on in isola on)
22 5/31/14
ARQ: Applica on-aware admission control
• Divide applica on workload into queues, using – Interference tolerance informa on – Heterogeneity requirement • Trade off between: (i) wai ng me; (ii) quality of a resource • Prevent highly demanding applica ons from blocking easy-to-sa sfy applica ons • Understand when a QoS viola on is “likely” – re-divert to a different queue • Interference func on (used to derive a resource quality): – Interference server can tolerate from the new applica on (c) – Interference new workload can tolerate from exis ng applica ons (t)
ARQ: Applica on-aware admission control
23 5/31/14
Disturbance Benchmarking
• Tolerance of an applica on to failure • Benchmark injects: – Workload & Disturbance into System Under Test – Measures response • Disturbance: – Events, faults, etc – Changes QoS profile of the applica on • Aim to measure “resilience” not availability – Approach similar to DBench-OLTP • Ability to adapt in the context of a disturbance in the system
From Aaron Brown and Peter Shum (IBM)
24 5/31/14
From Aaron Brown and Peter Shum (IBM)
Useful to compare this with performance benchmarks that we are much more aware of
Compare with automated tes ng mechanisms
From Aaron Brown and Peter Shum (IBM)
25 5/31/14
From Aaron Brown and Peter Shum (IBM)
From Aaron Brown and Peter Shum (IBM)
26 5/31/14
From Aaron Brown and Peter Shum (IBM)
From Aaron Brown and Peter Shum (IBM)
27 5/31/14
From Aaron Brown and Peter Shum (IBM)
Configura on Management • Dynamically deploy pre-configured virtual machine instances – Replicate across mul ple servers – Deploy a “reference” configura on across clients • CHEF – widely used configura on management tool (SaaS pla orm, Ruby-based) – Deploy load balancers, monitoring tools (Nagios) along with others (sharing “cookbooks” and “recipes”) – Apache Licence (with Apache SOLR (search engine), CouchDB) • CF Engine – Open source (GPL Licence) – Enables much more complex configura ons (-ve) – Uses a remote agent (also supports a monitoring deamon)
h p://www.slideshare.net/jeyg/configura on-manager-presenta on
28 5/31/14
Configura on Management
• Amazon CloudForma on another op on – Create & manage AWS instances -- h p://aws.amazon.com/cloudforma on/ – Provides pre-defined set of templates (WordPress, Joomla, Windows Server, Ruby on Rails, etc) -- h p://aws.amazon.com/cloudforma on/aws-cloudforma on-templates/ • CloudSo ’s Brooklyn h p://www.cloudso corp.com/communi es/ – Open source + support for policies – Applica on-level rather than instance-level support – Enables autonomic adapta on of a deployed configura on (e.g. auto-scaling policy, replacer/ restarter (high availability) policy)
Stream processing architectures • Systems that must react to streams of data produced by the external world • Stream data source can vary in complexity and type • Availability of streamed data can also be managed through an access control mechanism • Usually operate in real me over streams and generate in turns other streams of data enabling: (i) passive monitoring: what is happening, or (ii) ac ve control: sugges ng ac ons to perform, such as by stock X, raise alarm Y, or detected spa al viola on, etc. • Stream processing can also lead to seman c annota on of events
29 5/31/14
Difference from “standard” databases • Queries over streams are generally “con nuous” • execu ng for long periods of me • return incremental results • permanently installed – i.e. does not terminate a er first execu on • Performance metrics should be based on response me rather than comple on me • Data is not sta c – as new data is constantly arriving into the system – Same query at different mes leads to different results (as long as new data enters the system) • Typical opera ons in StreamSQL – SELECT (execute a func on on a stream) and WHERE (execute a filter on a stream) operators – Stream merge and join – Windowing and Aggrega on
Analyses of performance • Response Time – Average or maximum me between input arrival into the system, and the subsequent genera on of a response • Support (query) Load – What is the size of the input (number of data elements) a stream system can process while s ll mee ng specified response me target and correctness constraints • Correctness is me dependent – Same query at different mes à different outcomes – Poten ally mul ple correct answers depending on response me
30 5/31/14
Adaptive Streams • Three key issues: – what to remember or forget, – when to do the model update, and – how to do the model update • For streaming – these can be mapped into: – the size of the window to remember recent examples – methods for detec ng distribu on change in the input – methods for keeping updated es ma ons for input sta s cs
All “x” are real valued, estimator: current value of “x” + variance (each “x” independently drawn)
Estimator: linear, moving average, Kalman filter
Adaptive Sliding Windows (ADWIN) • Window size – reflects me scale of change – Small: reflects accurately the current distribu on – Large: many examples are available to work on, increasing accuracy in periods of stability • Window content is used for – detec ng change (e.g., by using some sta s cal test on different sub windows), – to obtain updated sta s cs from recent examples, – to have data to rebuild or revise the model(s) a er data has changed • Adap ve Windowing: – Whenever two “large enough” sub windows of W exhibit “dis nct enough” averages à corresponding expected values are different à drop older por on of window
31 5/31/14
Cloud-based stream processing • Use of Cloud resources to: – Execute stream processing operators (may be in- network) – VM per operator (dynamically allocated to overcome peak workloads) • Operator chaining within/across Cloud systems – Scale out – Fault tolerance • Operator chaining à processing pipelines – Similarity with workflow systems
GENI (OpenFlow and MiddleBox) • L2/L3 Technology to permit so ware-defined control of network forwarding and rou ng • Integra on of specialist network “appliances” to support specific func ons – These could be user defined – Linux hos ng • MiddleBox: In-network general-purpose processors fronted by OpenFlow switches • Integrate services from mul ple Clouds – Alloca on of networks and “slices” across different resources
32 5/31/14
In-transit Analysis
Delay (QoS parameter) D1 S1 T1 T2
T4 T5 S2 T3 D2
• Data processing while data is in movement from source to des na on • Ques on: what to process where and when • Use of “slack” in network to support par al processing • Applica on types: – Streaming & Data Fusion requirement
In-transit Analysis … 2
Shared Cluster D1 S1 T1 T2
T4 T5 S2 T3 D2
• Data processing while data is in movement from source to des na on • Ques on: what to process where and when • Use of “slack” in network to support par al processing
33 5/31/14
Workflow level Representa on [] [] ADSS [] [] [] PU1 (with in-transit PU2 processing)
Proc. Unit: t11,t12 ADSS: Proc. Unit: t21,t22,t23 t21,t22
mapping mapping
data transfer ADSS μ λ buffer B controller In transit δ nodes:
ω t21,t22 Resource r1: t11,t12 local storage Resource r2: t21,t22,t23 B Bandwidth Input rate λ ADSS model & Consumer’s data rate simulator δ μ Controlled output rate ω Disk transfer rate 70
Rafael Tolosana-Calasanz et al. “Revenue-based Resource Management on Shared Clouds for Heterogenous Bursty Data Streams”, GECON 2012, Springer
Approach & focus
Adap ve infrastructure for sensor data analysis
• Mul ple concurrent data streams with SLA • Variable proper es: rate and data types; various processing models • Support for in-transit analysis, enforcing QoS • Support for admission control & flow isola on at each node • In case of QoS viola on, penalisa on
A Scalable, Distributed Stream Processing System for Electric Vehicle Key focus Demand Forecas ng • Architectural components • Business rules for SLA Management : Ac ons to guarantee QoS & maximize revenue
From Jose Banares (University of Zaragoza)
34 5/31/14
From Jose Banares (University of Zaragoza) System Architecture
Data injec on rate SLA b R
• 3 key components / node: Token Bucket, Processing Unit & output streaming
From Jose Banares (University of Zaragoza) Token Bucket (shaping traffic)
slope Ra S(t) slope R A(t) slope R A(t) E(t) R tokens/s
b.C b C − R The real Traffic flow d tokens slope C B € a C bps time time A(t): Amount of data arriving up to time t Two key parameters of interest: • R: Also called the commi ed informa on rate (CIR), it specifies how much data can be sent or forwarded per unit me on average • B: it specifies for each burst how much data can be sent within a given me without crea ng scheduling concerns
35 5/31/14
Token Bucket (shaping traffic)
From Jose Banares (University of Zaragoza)
From Jose Banares (University of Zaragoza)
Token Bucket PU Buffer Input buffer
Each token bucket provides us tunable parameters: R,b
Controller: monitors & modifies behaviour
36 5/31/14
Resource addi on based on buffer occupancy
Autonomic Computa onal Science • Enable automated tuning of applica on behaviour – Execu on Units, Communica on, Coordina on, Execu on Environment – Rela on to “Reflec on” and Reflec ve Middleware + Use of intercession on a meta-model + domain-model – Developing a meta-model is o en difficult • Tuning may be: – Centralized – Consist of mul ple control units – Tuner external to the applica on • Comparison with Control systems & MDA – Mul ple, o en “hidden” control loops – Inclusion of run- me configura on parameters at design me – Model centric view that takes deployment and execu on into account
Shantenu Jha, Manish Parashar and Omer Rana, Self-adaptive architectures for autonomic computational science Proceedings of the First international conference on Self-organizing architectures , pp 177-197, LNCS 6090, Springer Verlag 2010.
37 5/31/14
Autonomic Computa onal Science Conceptual Framework
Tuning of applica on & resource manager parameters
Resources R Results Application Resource Inputs Manager R …
Tuning Tuning Parameters R Parameters
Autonomic Tuning
38 5/31/14
Tuning by applica on
Resources R Results Application Resource Inputs Manager R Tuning … Strategy Tuning Parameters R
Autonomic Tuning
Historical (monitoring) data
39 5/31/14
Spa al, Temporal and Computa onal Heterogeneity and Dynamics in SAMR
Spa al Heterogeneity
Temperature
Temporal Heterogeneity
Simula on of OH Profile combus on based on SAMR (H2-Air mixture; igni on via 3 hot-spots)
Courtesy: Sandia Na onal Lab
Autonomics in SAMR • Tuning by the applica on – Applica on level: when and where to refine – Run me/Middleware level: When, where, how to par on and load balance – Run me level: When, where, how to par on and load balance – Resource level: Allocate/de-allocate resources
• Tuning of the applica on, run me – When/where to refine – Latency aware ghost synchroniza on – Heterogeneity/Load-aware par oning and load-balancing – Checkpoint frequency – Asynchronous formula ons
40 5/31/14
Montage
Montage workflow
I/O intensive workflow
Significant data parallelism
41 5/31/14
Montage: Tuning Mechanisms
Montage: Tuning Strategy
42 5/31/14
Montage: Tuning Strategy
Montage: Tuning Strategy
43 5/31/14
Concluding comments • Autonomic strategies: – O en rooted in control systems (generally closed- loop feedback control) – Can use a variety of control strategies – which include use of machine learning • Formula ng the problem o en difficult – Mul -criteria op misa on – O en mul ple, difficult to separate control loops • Monitoring infrastructure choice is key
44