Responsive Interactive Applications by Dynamic Mapping of Activation Trees February 20, 1998

1
Responsive Interactive Applications by Dynamic
Mapping of Activation TreesFebruary 20, 1998

• Peter A. Dinda
• http//www.cs.cmu.edu/pdinda
• School of Computer Science
• Carnegie Mellon University

2
Outline

• Responsive interactive applications
• Best effort real-time service
• Dynamic mapping problem
• History-based prediction approach
• Statistical properties of host load
• Algorithms and evaluation for simplified problem
• Research plan
• Application and network traces

3
Interactive Application Model
Feedback
Message Handler
Message
mouse_click()
Aperiodic User Action
Activation tree
4
Acoustic Room Modeling
Room model
impulse responses
Physical Simulation of Wave Eqn
Speakers
Modify model
Frequency response plots
5
Other Applications

• Image editing
• The Adobe Photoshop universe
• Computer aided design
• Quake design optimization (Malcevic-97)
• Computational steering
• CUMULUS (Geist-96), CAVE (Disz-95),
• Games
• DIS
• Collaboration?
• Collaborative Planning (Zinky-DUTC-95)

6
Responsiveness

• Timely feedback to individual user actions
• Bound response time tmax
• Jitter bound and resource usage hint
• Bound response time ³ tmin
• Example image editor drawing tool

7
A Best Effort Real-time Service
MAP procedure() IN tmin, tmax

• Execute the activation tree rooted at procedure()
  so that tmintexectmax
• No guarantees
• Responsiveness spec bounds tmin,tmax
• Performance metric fraction of trees that meet
  their bounds

8
Machine Model

• Hosts on a LAN
• No centralized or coordinated scheduling, or
  reservations
• Other unrelated traffic exists
• We are only a user
• Remote execution facility
• Can execute any procedure on any host
• RPC, DSM, DCE, CORBA, DCOM, ...
• Measurable - at least a good real-time clock
  exists (lt1ms)

9
Execution Model
tmin,tmax

• Dynamically map nodes of the unfolding activation
  tree to the hosts
• At each procedure call, choose which host is best
  suited to execute the call in order to meet the
  bounds on the tree

10
Dynamic Mapping Problem

• How do we map the nodes of the trees to the hosts
  so that the fraction of trees that satisfy their
  bounds is maximized?

11
Aspects of My Approach

• History-based prediction
• Decomposition of bounds
• Adaptation of mapping algorithms during tree
  traversal
• Methods Trace-driven simulation,
• Iterative refinement

12
History-based Prediction
H0
foo()
tmin,tmax
...
tmin,tmax
H1
...
H0
bar()
...
foo() is executing on H0 and calls bar(), which
can be mapped to H0, H1, or H2
H2
H0 has a local history of execution times of
bar() on each of the other hosts

• For each host, H0 predicts whether it can meet
  the bounds, based on past local history and then
  chooses one where it is possible
• Execution times include both communication for
  remote call and the actual computation

13
Decomposition of Bounds
foo()
tmin,tmax
partially executed, known
tmin,tmax?
unexecuted, known
bar()
unexecuted, unknown
unexecuted, unknown

• Choice of tmin,tmax for bar() depends on
  unvisited portion of the tree
• Collect history of what fraction of time spent in
  foo() subtree was spent in bar() subtree
• Choose fraction of bounds to give to bar() based
  on that history and current time

14
Adaptation of Mapping Algorithms During Tree
Traversal

• Tune strategy to how deep we are in the tree and
  how far along in the traversal
• Explore more aggressively early in the traversal,
  when the effect of a bad decision is easiest to
  overcome
• Find interesting new hosts
• Spend less time making mapping decision deep in
  the tree
• More likely to remain on single host

15
Statistical Properties of Load

• Load traces
• First order properties of the traces
• Self-Similarity
• Epochal Behavior
• Goal modeling of load

16
Load Traces

• Digital Unix one minute load average
• 1 Hz sample rate, one week traces
• Wide variety of Hosts (38 machines)
• Production Cluster
• 13 machines in the PSCs Alpha Supercluster
• Research Cluster
• 8 machines in CMCL cluster
• Compute Servers
• 2 shared machines in CMCL lab
• Desktops
• 15 machines desktop workstations owned by CMCL
  members

17
Self-Similarity

• Intuition looks the same on all time scales
• Autocorrelations similar across time scales
• Power spectrum envelope decays differently from
  1/f
• Characterization Hurst parameter
• H Ranges from 0 to 1
• Hlt0.5 negative near neighbor correlation -
  choppy
• H0.5 no correlations
• Hgt0.5 positive near neighbor correlation -
  smooth
• Methodology Estimate Hurst Parameter
• Four different, validated, estimators

18
Hurst Parameter Estimates
19
What does Self-Similarity Tell Us?

• Load is complex, but not random
• History matters
• Smoothing may be misguided
• Load should frighten modelers
• Conventional stochastic process models are wrong
• Long memory stochastic process models are
  desirable but may be impractical

20
Epochal Behavior

• Local frequency content of load signal is stable
  for long periods of time with abrupt transitions
• Spectrogram shows wide vertical bands
• Not the same as seasonality

21
Epochal Behavior Example
22
Epoch Length Statistics
23
What Does EpochalBehavior Tell Us?

• Suggests decomposition of load trace modeling
  problem
• Segmentation problem to find epochs
• Modeling problem within each epoch

24
Algorithms and Evaluation for Simplified Problem

• Map only leaf nodes
• Ignore communication
• for I1 to N do MAP leaf_procedure() IN
  tmin,tmaxend

25
RangeCounter(W) A Near Optimal Algorithm

• Each host has a quality level Q and a window of
  the last W execution times (W is small)
• Choose host with highest quality level, and age
  quality levels of all hosts QQ-1
• If bounds are met, increase hosts quality level
  by the inverse of our confidence in it
• If bounds are not met, reduce hosts quality
  level by half QQ/2

26
Load Trace-based Simulation

• Exec time computed from load trace using a
  simple, validated model
• Mapping algorithms are given bounds, select a
  host, then are told exec time
• Simulator computes performance of
• Algorithm under test
• Optimal (precognizant) algorithm
• Random mapping
• Individual host mappings

27
Scope of Evaluation

• 9 mapping algorithms
• 6 different groups of hosts
• Chosen from 38 hosts
• 1 week, 1 Hz load trace from each host
• 648 different cases
• Combinations of nominal time and bounds
• 100,000 calls for each case

28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
Research Plan
Extend current results to the full dynamic
mapping problem

• Extend simulation environment to include
  communication and activation trees
• Trace collection (Activation trees, network)
• A trace for everything
• Trace characterization (models, benchmarks)
• Simulator extension
• Develop algorithm
• Evaluate with benchmarks
• Incorporate into real system

38
Activation Tree Traces

• Representation of real program runs
• Each node annotated with compute time, and what
  data it references
• Delay requirements

39
Network Traces

• Realistic communication times
• Packet traces on Ethernet with tcpdump
• Simple broadcast networks seem too limiting
• Remos
• Existing trace databases