Transcript Vivek Sarkar - UT Computer Science

IBM Research: Programming Technologies
Panel 1: Hot topics and future directions in
programming languages (PL) research
Vivek Sarkar, IBM Research
May 9, 2007
© 2007 IBM Corporation
My Background

Education
 B.Tech., IIT Kanpur, 1981 (Advisor: Keshav Nori)
 M.S., U Wisconsin-Madison, 1982
 Ph.D., Stanford University, 1987 (Advisor: John Hennessy)

Career at IBM
 1987 - 1990, PTRAN (Manager: Fran Allen)
 1991 - 1993, ASTI optimizer
 1994 - 1996, Application Development Technology Institute
 1997, MIT sabbatical
 1998 - 2001, Jalapeno / Jikes RVM
 2002 - present, PERCS (includes X10, Parallel Tools, Productivity)

Family
 Married with two daughters, 18 and 15
 Interests: Hiking, Theater, Horseback riding, Violin
2
PL Summer School, May 2007
© 2007 IBM Corporation
PL Research Opportunities: Examples
 Programming Models and Programming Language Design
 X10 (contact: Vijay Saraswat)
 Development Tools
 SAFARI (contact: Robert Fuhrer)
 Parallel Tools Platform (contact: Evelyn Duesterwald)
 Compilers, Managed Runtimes, Static & Dynamic Optimization
 Metronome (contact: David Bacon)
3
PL Summer School, May 2007
© 2007 IBM Corporation
X10 Vision: Portable Productive Parallel Programming
X10 Data Structures
X10 language defines
mapping from X10 objects
& activities to X10 places
X10 Places
X10 deployment defines
mapping from virtual X10
places to physical
processing elements
Physical PEs
Homogeneous
Multi-core
Heterogeneous
Accelerators
Clusters
SPE
PEs,
L1 $
...
PEs,
L1 $
SPU
...
SPU
SPU
SPU
SPU
SPU
SXU
SXU
SXU
SXU
SXU
SXU
LS
LS
LS
LS
LS
LS
LS
SMF
SMF
SMF
SMF
SMF
SMF
SMF
SMF
...
...
16B/cycle
PPE
PPU
L2
L1
SMP Node
MIC
16B/cycle (2x)
SMP Node
PEs,
PEs,
...
EIB (up to 96B/cycle)
16B/cycle
PEs,
L1 $
SPU
SXU
LS
16B/cycle
L2 Cache
PEs,
L1 $
SPU
SXU
Memory
...
PEs,
PEs,
...
Memory
BIC
PXU
32B/cycle 16B/cycle
L2 Cache
Dual
XDRTM
FlexIOTM
Interconnect
64-bit Power Architecture with VMX
4
PL Summer School, May 2007
© 2007 IBM Corporation
Overview of X10 (x10.sf.net)
Storage classes:

Activity-local

Place-local

Partitioned
global

Immutable
• Dynamic parallelism with a Partitioned Global Address Space
• Places encapsulate binding of activities and globally addressable data
• async (P) S --- run statement S asynchronously at place P
• finish S --- execute statement S, and wait for descendant async’s to terminate
• atomic S --- execute statement S atomically
• No place-remote accesses permitted in atomic section
Deadlock safety: any X10 program written with async,
atomic, and finish can never deadlock
5
PL Summer School, May 2007
© 2007 IBM Corporation
Java Grande Forum Example (Monte Carlo)
Multi-Threaded Java
public void runThread() {
Single-Threaded
Java
Distributed Multi-Threaded X10
results = new Vector(nRunsMC);
thobjects[] = new
initTasks() { tasksRunnable
= new ToTask[nRunsMC];
… }Runnable [JGFMonteCarloBench.nthreads];
initTasks() { tasks = new ToTask[dist.block([0:nRunsMC-1])]; … }
Thread th[] = new Thread [JGFMonteCarloBench.nthreads];
// Create
(nthreads-1) to share work public void runDistributed()
public void runSerial()
{
for(int i=1;i<JGFMonteCarloBench.nthreads;i++)
{
results = new Vector(nRunsMC);
{
thobjects[i] = new AppDemoThread(i,nRunsMC);
// Now do the computation.
results = new x10Vector(nRunsMC);
th[i] = new Thread(thobjects[i]);
PriceStock ps;
// Now do the computation
th[i].start();
for( int iRun=0; iRun
< nRunsMC; iRun++ ) {
finish ateach ( point[iRun] : tasks.distribution ) {
}
ps = new PriceStock();
PriceStock ps = new PriceStock();
// Parent thread acts as thread 0
ps.setInitAllTasks(initAllTasks);
ps.setInitAllTasks((ToInitAllTasks) initAllTasks);
thobjects[0] = new AppDemoThread(0,nRunsMC);
ps.setTask(tasks[iRun]);
ps.setTask(tasks[iRun]);
thobjects[0].run();
ps.run();
ps.run();
// Wait for child threads
results.addElement(ps.getResult());
final ToResult r = ps.getResult(); // ToResult is a value type
for(int i=1;i<JGFMonteCarloBench.nthreads;i++)
{
}
async(results)
atomic results.v.addElement(r);
try { th[i].join();} catch (InterruptedException
e) {}
}
}
}
}
}
class AppDemoThread implements Runnable {
Source: http://www.epcc.ed.ac.uk/javagrande/javag.html
... // initialization code - The Java Grande Forum Benchmark Suite
public void run() {
PriceStock ps;
int ilow, iupper, slice;
slice = (nRunsMC+JGFMonteCarloBench.nthreads-1)
/ JGFMonteCarloBench.nthreads;
ilow = id*slice;
iupper = Math.min((id+1)*slice, nRunsMC);
for( int iRun=ilow; iRun < iupper; iRun++ ) {
ps = new PriceStock();
ps.setInitAllTasks(AppDemo.initAllTasks);
ps.setTask(AppDemo.tasks[iRun]);
ps.run();
AppDemo.results.addElement(ps.getResult());
}
} // run()
}
6
PL Summer School, May 2007
© 2007 IBM Corporation
Lead: Robert Fuhrer
SAFARI Vision: Meta-Tooling for Language-Specific IDEs
 Problem
 Lack of tool support can be a significant barrier in adoption of new
languages
 SAFARI Solution: Meta-tools and framework
 Language generation tools (scanner/parser generator, high quality
automatic ASTs)
 Generation of Eclipse toolkit components
• Encapsulate Eclipse API knowledge
• Encapsulate common language structure, semantics, processing idioms
 Leverage language inheritance
•  structure/semantics   implementation
 People
 P. Charles, J. Dolby, R. Fuhrer, S. Sutton, M. Vaziri
7
PL Summer School, May 2007
© 2007 IBM Corporation
SAFARI Target IDE Functionality
syntax highlighting, compiler annotations,
hover help, source folding, formatting…
navigation (hyperlinks, “Open Type”, …)
content assist, quick fixes
structural views
compiler w/ incremental build,
automatic dependency tracking
analysis & refactoring
8

New Project/Type/… creation wizards

launch & debug: launch configs, breakpoints,
backtraces, values, evaluation
PL Summer School, May 2007
© 2007 IBM Corporation
Example of SAFARI Challenges: Error Handling
 Errors are the norm!  must not cripple the IDE!
mangled statement
A()
void A() {
int x= 5;
body
foo blah;
for(int i=0; i < a.length; i++) {
int x= 5;
BadStmt
int y= a[i] * a[j];
x += y;
}
header
}
dangling ref
int i=0;
i < a.length
for
body
i++
…
 SAFARI/LPG: systematic, semi-automatic error recovery for
parsing/creating “prosthetic” AST nodes
 Polyglot: ideas for finer-grained dependencies, better
robustness, make data dependencies more explicit
9
PL Summer School, May 2007
© 2007 IBM Corporation
Parallel Tools Platform Vision: Integrated Workbench for
High-Productivity Parallel Programming
PERCS workbench enhancements: MPI tools, OpenMP tools, Remote System Exploration, Performance
Exploration, Runtime Error Detection, Team Platform, Productivity measurements
HPC System
Parallel Tools
Platform (PTP)
Open HPC Workbench
(Runs on Windows,
Linux, Mac OS, …)
10
CPO
Static Analysis Tools
LL
PL Summer School, May 2007
HPC Toolkit
Cache injection
COE
ILM
Meiosys
ESSL
User Space
APPLICATION
PESSL
IBM’ s MPI
Remote interface
from Eclipse
Workbench to HPC
system
DB SMT Exploitation
SHMEM
UPC X10
LAPI
Kernel Space
GPFS
VSD/NSD
DD
HYP
HAL
SOCKETS
TCP
UDP
IP
Operating System
Compilers
CSM, RSCT
Eclipse
PTP is the
integration
hub for all
PERCS
tools
IF_LS
Network Adapter
HMC
Network
= New additions through PERCS to the HPC SW architecture
© 2007 IBM Corporation
PTP Example: MPI Barrier Verification Tool
Action to run
Barrier Verifier
Verify barrier
synchronization in C/MPI
programs
Synchronization errors lead
to deadlocks and stalls.
Programmers may have to
spend hours trying to find the
source of a deadlock
Static verification tools help
to eliminate errors before the
program is executed
Contact: Evelyn Duesterwald, Yuan Zhang
11
PL Summer School, May 2007
© 2007 IBM Corporation
MPI Barrier Verification Tool (contd.)
 MPI does not place any constraints
on the placement of barriers
MPI_Comm_rank(com, &rank)
rank > 2
potential deadlock
…
MPI_Barrier(com)
P(k)
i = rank
12
 Synchronization errors in MPI are
a common and difficult to find
problem
 MPI Barrier Verification:
Verify that the number of barriers
along concurrent paths is the same
…
i = F(0)
i>0
 Programmer has to ensure that the
number of barriers along concurrent
paths is the same
not a deadlock
MPI_Barrier(com)
PL Summer School, May 2007
- Match barriers that synchronize
- For unmatched barriers, report a
synchronization error with a counter
example that illustrates the error
© 2007 IBM Corporation
Metronome Vision:Transparent Real-time Java
C++ Application
Java Application
Java Application
Garbage
Collection
Java Runtime System
C++ Runtime System
(JVM)
Metronome
Java Runtime System
Manual, Unsafe
Automatic, Safe
Automatic, Safe
Predictable
Unpredictable
Predictable
www.research.ibm.com/metronome
13
PL Summer School, May 2007
© 2007 IBM Corporation
Real-time Garbage Collection


Garbage collection is fundamental to Java’s value proposition
 Safety, reliability, programmer productivity
 But also causes the most non-determinism (100 ms – 10 s latencies)
 RTSJ standard does not support use of garbage collection for real-time
Metronome is our hard real-time garbage collector
 Worst-case 2 ms latencies; high throughput and utilization
•
Research under way to further reduce real-time guarantee from ms to us
 100x better than competitors’ best garbage collection technology
Space
Time
Application Collector
Base Application Memory
14
Resulting Schedule
PL Summer School, May 2007
Garbage Collection Pause Times
(Customer application)
Worst-case 1.7 ms
Average 260 us
© 2007 IBM Corporation
Space g Time
Application (Mutator)
Scheduler
a*(∆GC) = Per-GC Allocate Rate
u = utilization
50%
75%
s = used space
45 MB
100 MB
∆t = time resolution
5 ms
15
PL Summer School, May 2007
50 MB/s
m = Live Data
30 MB
Collector
RT = Trace Rate
50 MB/s
RS = Sweep Rate
300 MB/s
© 2007 IBM Corporation
PL Research Opportunities
 Programming Models and Programming Language Design
 Drivers: Concurrency, Accelerators, Data Access, Web
Services, DSLs, …
 Development Tools
 Drivers: Program Analysis for Software Quality, Debugging
Tools, Performance Tools, Refactorings, Language-Sensitive
IDE’s, …
 Compilers, Managed Runtimes, Static & Dynamic Optimization
 Drivers: Hardware roadmap, PL trends, Virtualization,
Embedded systems, Real-time systems, …
16
PL Summer School, May 2007
© 2007 IBM Corporation
Additional Information
 X10, http://x10.sf.net
 SAFARI,
http://domino.research.ibm.com/comm/research_projects.nsf/pag
es/safari.index.html
 Parallel Tools platform, http://eclipse.org/ptp
 Metronome, http://www.research.ibm.com/metronome/
 IBM Research
 “Innovating at IBM” video,
http://www.research.ibm.com/about/career.shtml
 “Valuing diversity: an ongoing commitment”,
http://www.ibm.com/employment/us/diverse
17
PL Summer School, May 2007
© 2007 IBM Corporation