Transcript RENCISalsaOct22-07 - Community Grids Lab

Multicore SALSA
Parallel Computing and Web 2.0
for Cheminformatics and GIS Analysis
2007 Microsoft eScience Workshop at RENCI
The Friday Center for Continuing Education UNC - Chapel Hill
October 22 2007
Geoffrey Fox, Seung-Hee Bae, Neil Devadasan, Rajarshi Guha,
Marlon Pierce, Xiaohong Qiu, David Wild, Huapeng Yuan
Community Grids Laboratory, Research Computing UITS, School
of informatics and POLIS Center Indiana University
George Chrysanthakopoulos, Henrik Frystyk Nielsen
Microsoft Research, Redmond WA
http://www.infomall.org/multicore
[email protected], http://www.infomall.org
1
Too much Computing?



Historically one has tried to increase computing capabilities by
• Optimizing performance of codes
• Exploiting all possible CPU’s such as Graphics co-processors
and “idle cycles”
• Making central computers available such as NSF/DoE/DoD
supercomputer networks
Next Crisis in technology area will be the opposite problem –
commodity chips will be 32-128way parallel in 5 years time and
we currently have no idea how to use them – especially on clients
• Only 2 releases of standard software (e.g. Office) in this time
span
Gaming and Generalized decision support (data mining) are two
obvious ways of using these cycles
• Intel RMS analysis
• Note even cell phones will be multicore
Intel’s Projection
Too much Data to the Rescue?




Multicore servers have clear “universal parallelism” as many
users can access and use machines simultaneously
Maybe also need application parallelism as needed on client
machines
Over next years, we will be submerged of course in data
deluge
• Scientific observations for e-Science
• Local (video, environmental) sensors
• Data fetched from Internet defining users interests
Maybe data-mining of this “too much data” will use up the
“too much computing” both for science and commodity PC’s
• PC will use this data(-mining) to be intelligent user
assistant?
• Must have highly parallel algorithms
Intel’s Application Stack
CICC Chemical Informatics and Cyberinfrastructure
Collaboratory Web Service Infrastructure
Cheminformatics Services
Statistics Services
Database Services
Core functionality
Fingerprints
Similarity
Descriptors
2D diagrams
File format conversion
Computation functionality
Regression
Classification
Clustering
Sampling distributions
3D structures by
CID
SMARTS
3D Similarity
Docking scores/poses by
CID
SMARTS
Protein
Docking scores
Applications
Applications
Docking
Predictive models
Filtering
Feature selection
Druglikeness
2D plots
Toxicity predictions
Arbitrary R code (PkCell)
Mutagenecity predictions
PubChem related data by
Anti-cancer activity predictions
Need to make
Pharmacokinetic parameters
CID, SMARTS
all this parallel
OSCAR Document Analysis
InChI Generation/Search
Computational Chemistry (Gamess, Jaguar etc.)
Core Grid Services
Service Registry
Job Submission and Management
Local Clusters
IU Big Red, TeraGrid, Open Science Grid
Varuna.net
Quantum Chemistry
Portal Services
RSS Feeds
User Profiles
Collaboration as in Sakai
Deterministic Annealing for Data Mining




We are looking at deterministic annealing algorithms because
although heuristic
• They have clear scalable parallelism (e.g. use parallel BLAS)
• They avoid (some) local minima and regularize ill defined
problems in an intuitively clear fashion
• They are fast (no Monte Carlo)
• I understand them and Google Scholar likes them
Developed first by Durbin as Elastic Net for TSP
Extended by Rose (my student then; now at UCSB)) and Gurewitz
(visitor to C3P) at Caltech for signal processing and applied later to
many optimization and supervised and unsupervised learning
methods.
See K. Rose, "Deterministic Annealing for Clustering, Compression,
Classification, Regression, and Related Optimization Problems,"
Proceedings of the IEEE, vol. 80, pp. 2210-2239, November 1998
High Level Theory



Deterministic Annealing can be looked at from a
Physics, Statistics and/or Information theoretic point of
view
Consider a function (e.g. a likelihood) L({y}) that we
want to operate on (e.g. maximize)
Set L ({y},T) =  L({y}) exp(- ({y} - {y})2 /T ) d{y}
• Incorporating entropy term ensuring that one looks for most
likely states at temperature T
• If {y} is a distance, replacing L by L corresponds to smearing
or smoothing it over resolution T

Minimize Free Energy F = -Ln L ({y},T) rather than
energy E = -Ln L ({y})
• Use mean field approximation to avoid Monte Carlo
(simulated annealing)
Deterministic Annealing for Clustering I
N Points xi and K Cluster Centers yk
Pr( xi  Ck )  exp(  E ( xi , yk ) / T ) / Z ( xi ) where
Z ( xi )  k 1 exp(  E ( xi , yk ) / T )
K
E ( xi , yk )  ( xi  yk ) 2
Free Energy F  T i 1 ln Z ( xi ))
N
Compare Simple Gaussian Mixture (K centers) with
Z ( xi )  k 1 Pk exp(  E ( xi , yk ) /( 2 k ))
K


2
Illustrating similarity between clustering and Gaussian mixtures
Deterministic annealing for mixtures replaces 2 k 2 by 2 k 2  T
and anneals down to mixture size
Deterministic Annealing for Clustering II
with Pr( xi  Ck )  exp(  E ( x i , y k
Calculate y k
new

N
i 1
old
) / T ) / Z ( xi , y k
x i Pr( x i  Ck )

N
i 1






old
)
Pr( x i  Ck )
This is an extended K-means algorithm
Start with a single cluster giving as solution y1 as centroid
For some annealing schedule for T, iterate above algorithm
testing correlation matrix in xi about each cluster center to see if
“elongated”
Split cluster if elongation “long enough”; splitting is a phase
transition in physics view
You do not need to assume number of clusters but rather a final
resolution T or equivalent
At T=0, uninteresting solution is N clusters; one at each point xi
Deterministic
Annealing
F({y}, T)
Solve Linear
Equations for
each temperature
Nonlinearity
removed by
approximating
with solution at
previous higher
temperature


Configuration {y}
Minimum evolving as temperature decreases
Movement at fixed temperature going to local
minima if not initialized “correctly
Clustering Data



Cheminformatics was tested successfully with small datasets and
compared to commercial tools
Cluster on properties of chemicals from high throughput
screening results to chemical properties (structure, molecular
weight etc.)
Applying to PubChem (and commercial databases) that have 620 million compounds
• Comparing traditional fingerprint (binary properties) with real-valued
properties

GIS uses publicly available Census data; in particular the 2000
Census aggregated in 200,000 Census Blocks covering Indiana
• 100MB of data

Initial clustering done on simple attributes given in this data
• Total population and number of Asian, Hispanic and Renters

Working with POLIS Center at Indianapolis on clustering of
SAVI (Social Assets and Vulnerabilities Indicators) attributes at
http://www.savi.org) for community and decision makers
• Economy, Loans, Crime, Religion etc.
Where are we?


We have deterministically annealed clustering running well on 8core (2-processor quad core) Intel systems using C# and
Microsoft Robotics Studio CCR/DSS
Could also run on multicore-based parallel machines but didn’t
do this (is there a large Windows quad core cluster on
TeraGrid?)
• This would also be efficient on large problems

Applied to Geographical Information Systems (GIS) and census
data
• Could be an interesting application on future broadly deployed PC’s
• Visualize nicely on Google Maps (and presumably Microsoft Virtual Earth)


Applied to several Cheminformatics problems and have parallel
efficiency but visualization harder as in 150-1024 (or more)
dimensions
Will develop a family of such parallel annealing data-mining
tools where basic approach known for
• Clustering
• Gaussian Mixtures (Expectation Maximization)
• and possibly Hidden Markov Methods
Clustering algorithm annealing by decreasing distance scale and gradually finds more
clusters as resolution improved
Here we see 10 clusters increasing to 30 as algorithm progresses
Total
Total
Asian
Asian
Hispanic
Hispanic
Purdue
Renters
Renters
Renters
IUB
30 Clusters
10 Clusters
In detail, different groups have
different cluster centers
Multicore SALSA at CGL

Service Aggregated Linked Sequential Activities
• http://www.infomall.org/multicore

Aims to link parallel and distributed (Grid) computing
by developing parallel applications as services and not
as programs or libraries
• Improve traditionally poor parallel programming
development environments

Can use messaging to link parallel and Grid services
but performance – functionality tradeoffs different
• Parallelism needs few µs latency for message latency and
thread spawning
• Network overheads in Grid 10-100’s µs

This presentation describes first of set of services
(library) of multicore parallel data mining algorithms
Parallel Programming Model







If multicore technology is to succeed, mere mortals must be able to
build effective parallel programs
There are interesting new developments – especially the Darpa HPCS
Languages X10, Chapel and Fortress
However if mortals are to program the 64-256 core chips expected in 5-7
years, then we must use today’s technology and we must make it easy
• This rules out radical new approaches such as new languages
The important applications are not scientific computing but most of the
algorithms needed are similar to those explored in scientific parallel
computing
• Intel RMS analysis
We can divide problem into two parts:
• High Performance scalable (in number of cores) parallel kernels or
libraries
• Composition of kernels into complete applications
We currently assume that the kernels of the scalable parallel
algorithms/applications/libraries will be built by experts with a
Broader group of programmers (mere mortals) composing library
members into complete applications.
Scalable Parallel Components





There are no agreed high-level programming environments for
building library members that are broadly applicable.
However lower level approaches where experts define
parallelism explicitly are available and have clear performance
models.
These include MPI for messaging or just locks within a single
shared memory.
There are several patterns to support here including the
collective synchronization of MPI, dynamic irregular thread
parallelism needed in search algorithms, and more specialized
cases like discrete event simulation.
We use Microsoft CCR
http://msdn.microsoft.com/robotics/ as it supports both MPI
and dynamic threading style of parallelism
Composition of Parallel Components










The composition step has many excellent solutions as this does not
have the same drastic synchronization and correctness constraints as
for scalable kernels
• Unlike kernel step which has no very good solutions
Task parallelism in languages such as C++, C#, Java and Fortran90;
General scripting languages like PHP Perl Python
Domain specific environments like Matlab and Mathematica
Functional Languages like MapReduce, F#
HeNCE, AVS and Khoros from the past and CCA from DoE
Web Service/Grid Workflow like Taverna, Kepler, InforSense KDE,
Pipeline Pilot (from SciTegic) and the LEAD environment built at
Indiana University.
Web solutions like Mash-ups and DSS
Many scientific applications use MPI for the coarse grain composition
as well as fine grain parallelism but this doesn’t seem elegant
The new languages from Darpa’s HPCS program support task
parallelism (composition of parallel components) decoupling
composition and scalable parallelism will remain popular and must be
supported.
“Service Aggregation” in SALSA





Kernels and Composition must be supported both inside
chips (the multicore problem) and between machines in
clusters (the traditional parallel computing problem) or
Grids.
The scalable parallelism (kernel) problem is typically only
interesting on true parallel computers as the algorithms
require low communication latency.
However composition is similar in both parallel and
distributed scenarios and it seems useful to allow the use of
Grid and Web 2.0 composition tools for the parallel problem.
• This should allow parallel computing to exploit large
investment in service programming environments
Thus in SALSA we express parallel kernels not as traditional
libraries but as (some variant of) services so they can be used
by non expert programmers
For parallelism expressed in CCR, DSS represents the
natural service (composition) model.
Inside the SALSA Services




We generalize the well known CSP (Communicating
Sequential Processes) of Hoare to describe the low level
approaches to fine grain parallelism as “Linked Sequential
Activities” in SALSA.
We use term “activities” in SALSA to allow one to build
services from either threads, processes (usual MPI choice)
or even just other services.
We choose term “linkage” in SALSA to denote the different
ways of synchronizing the parallel activities that may
involve shared memory rather than some form of messaging
or communication.
There are several engineering and research issues for
SALSA
• There is the critical communication optimization
problem area for communication inside chips, clusters
and Grids.
• We need to discuss what we mean by services
Microsoft CCR
• Supports exchange of messages between threads using named
ports
• FromHandler: Spawn threads without reading ports
• Receive: Each handler reads one item from a single port
• MultipleItemReceive: Each handler reads a prescribed number of
items of a given type from a given port. Note items in a port can
be general structures but all must have same type.
• MultiplePortReceive: Each handler reads a one item of a given
type from multiple ports.
• JoinedReceive: Each handler reads one item from each of two
ports. The items can be of different type.
• Choice: Execute a choice of two or more port-handler pairings
• Interleave: Consists of a set of arbiters (port -- handler pairs) of 3
types that are Concurrent, Exclusive or Teardown (called at end
for clean up). Concurrent arbiters are run concurrently but
exclusive handlers are
• http://msdn.microsoft.com/robotics/
23
MPI Exchange Latency in µs (20-30 µs computation between messaging)
Machine
OS
Runtime
Grains
Parallelism
MPI Exchange
Latency
Intel8c:gf12
(8 core 2.33 Ghz)
(in 2 chips)
Redhat
MPJE (Java)
Process
8
181
MPICH2 (C)
Process
8
40.0
MPICH2: Fast
Process
8
39.3
Nemesis
Process
8
4.21
MPJE
Process
8
157
mpiJava
Process
8
111
MPICH2
Process
8
64.2
Vista
MPJE
Process
8
170
Fedora
MPJE
Process
8
142
Fedora
mpiJava
Process
8
100
Vista
CCR (C#)
Thread
8
20.2
XP
MPJE
Process
4
185
Redhat
MPJE
Process
4
152
mpiJava
Process
4
99.4
MPICH2
Process
4
39.3
XP
CCR
Thread
4
16.3
XP
CCR
Thread
4
25.8
Intel8c:gf20
(8 core 2.33 Ghz)
Intel8b
(8 core 2.66 Ghz)
AMD4
(4 core 2.19 Ghz)
Intel4 (4 core 2.8 Ghz)
Fedora
Preliminary Results
• Parallel Deterministic Annealing Clustering in
C# with speed-up of 7.8 (Chemistry) and 7
(GIS) on Intel 2 quad core systems
• Analysis of performance of Java, C, C# in
MPI and dynamic threading with XP, Vista,
Windows Server, Fedora, Redhat on
Intel/AMD systems
• Study of cache effects coming with MPI
thread-based parallelism
• Study of execution time fluctuations in
Windows (limiting speed-up to < 8)
DSS as Service Model
• We view system as a collection of services
– in this case
– One to supply data
– One to run parallel clustering
– One to visualize results – in this by spawning
a Google maps browser
– Note we are clustering Indiana census data
• DSS is convenient as built on CCR
• Messaging overhead around 30-40 µs
Parallel Multicore GIS
Deterministic Annealing Clustering
Parallel Overhead
on 8 Threads Intel 8b
0.45
10 Clusters
0.4
Overhead = Constant1 + Constant2/n
Speedup = 8/(1+Overhead)
0.35
Constant1 = 0.02 to 0.1 (Client Windows) due to thread
runtime fluctuations
0.3
0.25
20 Clusters
0.2
0.15
0.1
0.05
10000/(Grain Size n = points per core)
0
0
0.5
1
1.5
2
2.5
3
3.5
4
Parallel Multicore
Deterministic Annealing Clustering
Parallel Overhead for large (2M points) Indiana Census clustering
on 8 Threads Intel 8b
This fluctuating overhead due to 5-10% runtime fluctuations between threads
0.250
0.200
overhead
“Constant1”
0.150
0.100
0.050
Increasing number of clusters decreases
communication/memory bandwidth overheads
0.000
0
5
10
15
20
#cluster
25
30
35
Parallel Multicore
Deterministic Annealing Clustering
0.200
Parallel Overhead for subset of PubChem clustering on 8 Threads
(Intel 8b)
0.180
“Constant1”
The fluctuating overhead
is reduced to 2% (under investigation!)
40,000 points with 1052 binary properties
(Census is 2 real valued properties)
0.160
overhead
0.140
0.120
0.100
0.080
0.060
0.040
Increasing number of clusters decreases
communication/memory bandwidth overheads
0.020
0.000
0
2
4
6
8
10
#cluster
12
14
16
18
MPI Parallel Divkmeans clustering of PubChem
700
min_size
650
600
Runtime (seconds)
ncpus
wall_mins
1
1
1
1
100
100
100
100
100
1000
1000
1000
1000
550
500
450
20
40
60
80
20
40
40
60
80
20
40
60
80
676
444
379
353
462
356
356
339
337
513
376
346
346
walltime
11:16:06
7:24:24
6:18:41
5:53:00
7:41:58
5:56:01
5:55:47
5:38:44
5:36:53
8:32:39
6:16:25
5:46:22
5:45:40
400
350
300
250
0
10
20
30
40
50
60
Number of processors
Minsize 1
Minsize 100
AVIDD Linux cluster, 5,273,852 structures
(Pubchem compound collection, Nov 2005)
Minsize 1000
70
80
90
Scaled Speed up Tests
• The full clustering algorithm involves different values of the
number of clusters NC as computation progresses
• The amount of computation per data point is proportional to NC
and so overhead due to memory bandwidth (cache misses)
declines as NC increases
• We did a set of tests on the clustering kernel with fixed NC
• Further we adopted the scaled speed-up approach looking at
the performance as a function of number of parallel threads
with constant number of data points assigned to each thread
– This contrasts with fixed problem size scenario where the number of data
points per thread is inversely proportional to number of threads
• We plot Run time for same workload per thread divided by
number of data points multiplied by number of clusters multiped
by time at smallest data set (10,000 data points per thread)
• Expect this normalized run time to be independent of number of
threads if not for parallel and memory bandwidth overheads
– It will decrease as NC increases as number of computations per points
fetched from memory increases proportional to NC
Intel 8-core C# with 80 Clusters: Vista Run
Time Fluctuations for Clustering Kernel
• 2 Quadcore Processors
80 Cluster(ratio
of std to timeofvsrun
#thread)
• This is average of standard
deviation
time of the 8 threads
between messaging synchronization points
0.1
Standard Deviation/Run Time
10,000 Datpts
50,000 Datapts
0.05
500,000 Datapts
Number of Threads
0
0
1
2
3
4
5
6
7
8
Intel 8 core with 80 Clusters: Redhat Run
Time Fluctuations for Clustering Kernel
• This is average of standard deviation of run time of the
80 Cluster(ratio of std to time vs #thread)
8 threads between messaging synchronization points
0.006
Standard Deviation/Run Time
0.004
10,000 Datapts
50,000 Datapts
0.002
500,000 Datapts
Number of Threads
0
1
2
3
4
5
6
7
8
Basic Performance of CCR
CCR Overhead for a computation of
23.76 µs between messaging
Intel8b: 8 Core
(μs)
Pipeline
Spawned
Rendez
vous
MPI
Number of Parallel Computations
1
1.58
2
2.44
3
3
4
2.94
7
4.5
8
5.06
Shift
2.42
3.2
3.38
5.26
5.14
Two Shifts
Pipeline
4.94
3.96
5.9
4.52
6.84
5.78
14.32 19.44
6.82 7.18
Shift
Exchange As
Two Shifts
4.46
6.42
5.86
10.86 11.74
7.4
11.64
14.16 31.86 35.62
Exchange
6.94
11.22
13.3
2.48
18.78 20.16
30
Time Microseconds
AMD Exch
25
AMD Exch as 2 Shifts
AMD Shift
20
15
10
5
Stages (millions)
0
0
2
4
6
8
10
Overhead (latency) of AMD4 PC with 4 execution threads on MPI style Rendezvous
Messaging for Shift and Exchange implemented either as two shifts or as custom CCR
pattern
70
Time Microseconds
60
Intel Exch
50
Intel Exch as 2 Shifts
Intel Shift
40
30
20
10
Stages (millions)
0
0
2
4
6
8
10
Overhead (latency) of Intel8b PC with 8 execution threads on MPI style Rendezvous
Messaging for Shift and Exchange implemented either as two shifts or as custom
CCR pattern
Cache Line Interference
•
•
•
•
•
Cache
Line
Interference
Early implementations of our clustering algorithm
showed large fluctuations due to the cache line
interference effect discussed here and on next slide
in a simple case
We have one thread on each core each calculating a
sum of same complexity storing result in a common
array A with different cores using different array
locations
Thread i stores sum in A(i) is separation 1 – no
variable access interference but cache line
interference
Thread i stores sum in A(X*i) is separation X
Serious degradation if X < 8 (64 bytes) with Windows
– Note A is a double (8 bytes)
– Less interference effect with Linux – especially Red Hat
Cache Line Interference
•
•
•
Machine
OS
Run
Time
Intel8b
Intel8b
Intel8b
Intel8b
Intel8a
Intel8a
Intel8a
Intel8c
AMD4
AMD4
AMD4
AMD4
AMD4
AMD4
Vista
Vista
Vista
Fedora
XP CCR
XP Locks
XP
Red Hat
WinSrvr
WinSrvr
WinSrvr
XP
XP
XP
C# CCR
C# Locks
C
C
C#
C#
C
C
C# CCR
C# Locks
C
C# CCR
C# Locks
C
Time µs versus Thread Array Separation (unit is 8 bytes)
1
4
8
1024
Mean Std/
Mean
Std/
Mean Std/
Mean Std/
Mean
Mean
Mean
Mean
8.03
.029
3.04
.059
0.884 .0051
0.884 .0069
13.0
.0095 3.08
.0028
0.883 .0043
0.883 .0036
13.4
.0047 1.69
.0026
0.66
.029
0.659 .0057
1.50
.01
0.69
.21
0.307 .0045
0.307 .016
10.6
.033
4.16
.041
1.27
.051
1.43
.049
16.6
.016
4.31
.0067
1.27
.066
1.27
.054
16.9
.0016 2.27
.0042
0.946 .056
0.946 .058
0.441 .0035 0.423
.0031
0.423 .0030
0.423 .032
8.58
.0080 2.62
.081
0.839 .0031
0.838 .0031
8.72
.0036 2.42
0.01
0.836 .0016
0.836 .0013
5.65
.020
2.69
.0060
1.05
.0013
1.05
.0014
8.05
0.010
2.84
0.077
0.84
0.040
0.840 0.022
8.21
0.006
2.57
0.016
0.84
0.007
0.84
0.007
6.10
0.026
2.95
0.017
1.05
0.019
1.05
0.017
Note measurements at a separation X of 8 (and values between 8 and 1024 not shown)
are essentially identical
Measurements at 7 (not shown) are higher than that at 8 (except for Red Hat which
shows essentially no enhancement at X<8)
If effects due to co-location of thread variables in a 64 byte cache line, the array must be
aligned with cache boundaries
–
In early implementations we found poor X=8 performance expected in words of A split across
cache lines
Inter-Service Communication

Note that we are not assuming a uniform
implementation of service composition even if user sees
same interface for multicore and a Grid
• Good service composition inside a multicore chip can require
highly optimized communication mechanisms between the
services that minimize memory bandwidth use.
• Between systems interoperability could motivate very
different mechanisms to integrate services.
• Need both MPI/CCR level and Service/DSS level
communication optimization

Note bandwidth and latency requirements reduce as
one increases the grain size of services
• Suggests the smaller services inside closely coupled cores and
machines will have stringent communication requirements.
Mashups v Workflow?


Mashup Tools are reviewed at
http://blogs.zdnet.com/Hinchcliffe/?p=63
Workflow Tools are reviewed by Gannon and Fox
http://grids.ucs.indiana.edu/ptliupages/publications/Workflow-overview.pdf



Both include scripting
in PHP, Python, sh etc.
as both implement
distributed
programming at level
of services
Mashups use all types
of service interfaces
and perhaps do not
have the potential
robustness (security) of
Grid service approach
Mashups typically
“pure” HTTP (REST)
43