Machine Learning and Grid Optimization

1
Machine Learning and Grid Optimization

• Noah Edelson
• Distributed Systems Department

2
Motivation

• Computation is important to science.
• Search is one of the most important problems in
  computation.
• Uninformed blind search is the most general
  search technique.
• Uninformed searches often perform better than
  uninformed programmers.

3
The Black Art of Fussing with Parameters

• gt Rather than cranking the buffers all the way
  up, it may be better to
• gt use multiple streams to fill the pipe. It may
  not possible to saturate
• gt a a fat pipe with just one connection for the
  above reasons, but also
• gt because of traffic shapers and various other
  limits. You can always
• gt try of course -) For parallel connections use
  some reasonable
• gt estimate, e.g. divide the maximum "pipe width"
  (bandwith rtt) by what
• gt you seem to be able to get out from a single
  stream, give or take a few
• gt more streams. It would probably help to be
  multi-threaded if using
• gt multiple streams (to avoid txqueuelen
  exhaustion). Alternatively,
• gt simply copying several files in parallel
  achieves the same.

4
Blind Search

• Queue lt- MakeQueue(Initial-State)
• while( Not Empty(Queue) )
• Node lt- Remove-Front(Queue)
• if( Goal-Test(Node) )
• return Solution
• Queue lt- Queuing-FN( Expand(Node), Queue)
• return Failure

5
Search Taxonomy
6
Evolutionary Computation

• .

Selection
Recombination (crossover)
Mutation
Replacement
Borrowed from Alba Troya - Improving
flexibility and efficiency by adding parallelism
to genetic algorithms, Statistics and Computing 12
7
Genetic Algorithms
t 1
t
reproduction
selection
Borrowed from Alba Troya - Improving
flexibility and efficiency by adding parallelism
to genetic algorithms, Statistics and Computing 12
8
Combinatorial Explosion

• A search, by definition, picks 1 of N solutions
  from the solution space.
• Given a bitstring with cardinality N, there are
  2N possible states to search.
• A well known AI professor at Cal claims that GA
  ignore the combinatoral explosion of the search
  space. He is wrong.

9
The Black Art of Fitness Functions

• A good fitness function converges quickly.
• Minimize genes (degrees of freedom)
• Avoid Deceptive Trap Functions
• Encourage smooth fitness functions
• Decouple fitness dependence between genes
• -- What happens when your fitness function halts?

10
A Solution to the Entscheidungsproblem!!!

• When all else fails- give up!
• A solution whose evaluation halts is decidedly
  unfit.
• Boils down to throwing out the outliers

11
Parameter Tuning by way of Search (2 steps!)

• An optimizer simply performs a search for
  optimal performance.
• Step 1) Encode your input parameters
• Step 2) Invent a fitness metric-
• I used ExecutionTime-1

12
Example Fitness Function

• def testFunc(data, args'')
• if not data return 3, 1.0, 0.0
• p,b,t int(1data010.0) , \
• 81024int(int(data1255.0)
  )1024, \
• 81024int(int(data2255.0)
  1024)
• startTime time.time()
• for samples in range(num_samples)
• lazy_forker(globus-url-copy ' -p '
  str(p)' -bs ' str(b) ' -tcp-bs\
• str(t) ' ' src ' ' dst, threshHold)
• endTime time.time() - startTime
• return 1.0/endTime

13
Step 3?

• Step 3) Grin in satisfaction, as your GA will
  explore multiple dimensions of your
  solution-space at once.

14
Results

• Globus-url-copy pyGlobus-url-copy
• (2x avg) (4x avg)
• Performance Boost
• 64 35 19
• 128 22 15
• 192 08 00
• 256 03 06
• mb

15
Results

• Globus-url-copy and pyGlobus-url-copy are similar
  in terms of performance given input parameters
  and running environment.
• Parameter tuning affords both utilities a
  noticeable increase in performance given little
  knowledge of the problem domain.

16
Why an optimizer service?

• .

Optimizer TuneService(FF) TestEnv
Save/Load TuningsEnv
FitnessFunction(x,y,z)
FitnessFunction(x,y,z)
GridServiceA(x,y,z)
GridServiceB(w,q,r)
17
Too Much Work?

• Writing an application-tuning fitness function is
  similar to wrapping or service-ifying the
  application.
• If you wrap a well-used app whose performance is
  sensitive to the parameters you support, you
  might as well provide a mechanism for tuning
  those parameters for higher performance.

18
Future Work

• Service-ify the GA
• - What standard to use?
• Production runs on WAN?
• - PlanetLab, EMULab?
• What besides pyGlobus-url-copy and
  globus-url-copy?
• - Grid Schedulers?
• - RLS Strategies?
• - Parameter tuning for important applications
  services?

19
Thanks!

• Have a good weekend!