From EARTH to HTMT: The Evolution of a Multithreaded Architecture Model

1
From EARTH to HTMTThe Evolution of a
Multithreaded Architecture Model

• Guang R. Gao
• Computer Architecture Parallel Systems
  Laboratory
• (CAPSL)
• University of Delaware

2
Outline

• Introduction
• The EARTH Execution and Architecture Model
• The EARTH Programming Model and Threaded-C
• Application Studies and Performance Evaluation
• Related Work and Conclusions

3
Scalable

• Main Challenges
• High-Performance
• Parallel Systems

for both Class A and Class B Applications
4
Challenges The Killer Latency Problem
P
Latency due to - Communication -
Synchronization - task spawning - load balancing
NI
C
M
Network
P
NI
C
M
SP2 is hard enough, PC clusters is much worse !
5
Meeting High-End Application Challenges

• Observation I Many such Applications have Bad
  Latencies demanding good support of adaptive
  fine-grain parallelism

Petaflop-2 Conference, 99-2
6
Here Comes the Surprise!Theobalds Ph.D.
thesis, May, 1999

• Observation II It is not necessarily too hard to
  generate and program fine-grain threads!

However, it may be hard to statically group them
into coarse-grain threads!
7
A Base Adaptive Fine-Grain Multithreaded
Execution Model
C1 (Abundance) a very large pool of threads
C2 (ultra-light weight) can be spawned as
easily and as quickly as possible
C3 (Mobility) Adaptively migratable as easily
and as quickly as possible
8
Motivation of The EARTH Project
How to exploit fine-grain multithreadeding on a
parallel system given off-the-shelf
microprocessors
9
Two Types of Fine-Grain Threads

• A parallel function invocation
• Strand/Fiber - a function body can be divided
  into several strands/fibers

10
Threads and Fibers

• A fiber becomes enabled if it has received all
  input signals
• An enabled fiber may be selected for
• execution when the required hardware
• resource has been allocated

• After finished execution, a signal is sent to all
  destination fiber to update the corresponding
  sync slots

Note The role of strand !
11
The Execution Model of Fibers

• Dependence-Driven firing rule for fibers
• Fiber is atomic and ultra-light weighted
• Relation with dataflow model (Dennis72)

12
The Threaded C Language

• Threaded C ANSI C extensions for
  multithreading
• Extensions include
• Threaded functions
• Threaded synchronization
• Support for global addresses
• Data transfer primitives
• Threaded C is
• The instruction set of the
• EARTH processor
• A target language for
• high-level compilers

C
FORTRAN
High-Level Language Translation
Users
Treaded C
Threaded C Compiler
EARTH Platforms
13
(No Transcript)
14
An Evolutionary Path for EARTH
SU-int
SU-ext
CPU / SU
CPU
SU
MANNA-dual/spn
- Parallel machines - PC-clusters - ...
lt
SEMi Simulation Platform (Theobald99)
15
Platforms for EARTH

• MANNA
• MANNA is architecture testbed from GMD
• benchmarking platform for fine-grain
  multithreading
• EARTH-SP2
• EARTH-Beowulf (Linux based)
• EARTH-SUN/SMP/Cluster

16
Unique Advantages of EARTH-MANNA Platform

• We can push OS completely out of the way!
• We can design the EARTH runtime system from very
  low level up
• The invaluable experience/lessons learned from
  EARTH-MANNA are essential for the successful
  migration of the EARTH model to other platforms
  (e.g. the IBM SP-2 story, etc.)

17
(No Transcript)
18
(No Transcript)
19
Performance of EARTH-MANNAon N-Queens(12)(1,637,
099 tokens are generated! On the average, 30
tokens are maintained per processors)
Parafin, protein folding, etc.
20
Sumamry of Recent Experimental Results (Kevin99)

• Impressive speedup and scalability (scalable even
  with high overhead fine-grain parallel programs
  e.g. fib)
• Enhanced Programmability (N-queen-p example)
• Broad applicability

21
Experiements

• Example 1 (assorted benchmarks) fib, nqueen,
  paraffin, tomcatv, matrix-multiply,etc.
• Example 2 Adaptive unstructured grids
• Example 3 Wavelet computation

22
(No Transcript)
23
(No Transcript)
24
Performance of N-Queens(12)Theobald99

• 117.8 fold speedup on a 120 node simulation!
• 1,637,099 tokens are generated !
• average, 30 tokens are maintained per processors
• n-QUUEN is a useful HTMT benchmark after all !
  (Phil Murkey)

25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
Coarse-Grain Applications

• 116 fold speedup on 120-node machine is achieved
  for Cannons matrix multiply algorithm!
• Deep software systolic-style implementation to
  exploit paralelism
• Fine-grain mechanisms

29
Example 2 --- Adaptive Unstructured Mesh
Computation

• Observation
• The critical part of the framework is mesh
  adaptation and load balancing
• Partitioning problem in better shape, remapping
  problem open

30
(No Transcript)
31
Initial Results

• About 3000 lines of Threaded-C code
• migration gt 70 (good)
• Unbiased variance 3 - 5 (very good)
• A good speedup on EARTH-MANNA has been observed

32
Performance of 2D Mesh Adaptation
EARTH-MANNA vs SP-2
EARTH-MANNA SP-2
33
Example 3 --- Adaptive Wavelet Transformation

• Load evolution pattern is dynamically changing,
  but is statically predictable
• Need adaptive load redistribution/grouping
• Mapping onto EARTH IPPS99

34
HTMT Facility (Perspective)
35
HTMT Architecture
36
Extensions to CurrentEARTH Model

• Percolation Model
• Memory Model Location Consistency
• Load balancing and percolation

37
HTMT Percolation Model
CRYOGENIC AREA
I-Pool
A-Pool
Parcel Invocation Termination
Parcel Assembly Disassembly
Parcel Dispatcher Dispenser
T-Pool
D-Pool
Run Time System
DMA
SRAM-PIM
38
The System Software Architecture

• Note
• The threaded-C compiler has part of its functions
  embedded in RTS
• The RTS will work with architecture and OS layers
  to provide the PXM interface
• The performance models Are defined across all
  layers

39
Evolution of Multithreaded Architecture Models
CHoPP77
CHoPP87
Non-dataflow based
MASA Halstead 1986
Alwife Agarwal 1989-96
XMT Vishkin
HEP B. Smith 1978
Tera B. Smith 1990-
CDC 6600 1964
J-Machine Dally 1988-93
M-Machine Dally 1994-98
Flynns Processor 1969
Cosmic Cube Seiltz 1985
Others Multiscalar (1994), SMT (1995), etc.
Monsoon Papadopoulos Culler 1988
Dataflow model inspired
T/Start-NG MIT/Motorola 1991-
P-RISC Nikhil Arvind 1989
MIT TTDA Arvind 1980
Cilk Leiserson
TAM Culler 1990
Iannucis 1988-92
Static Dataflow Dennis 1972 MIT
EM-5/4/X RWC-1 1992-97
Manchester Gurd Watson 1982
SIGMA-I Shimada 1988
Arg-Fetching Dataflow DennisGao 1987-88
MTA HumTheobald Gao 94
EARTH PACT95, ISCA96, Theobald99
MDFA Gao 1989-93
40
Acknowledgement(Incomplete List)
NSERC, FCAR, DARPA,NSA,NSF,NASA

• Erik Altman
• Haiying Cai
• Nasser Elmasri
• Gerd Heber
• Laurie J. Hendren
• Herbert Hum
• Alberto Jimenez
• Prasad Kakulavarapu
• Cheng Li
• Olivier Maquelin
• Andres Marquez

• Shashank Nemawarkar
• Zach Ruiz
• Sean Ryan
• V.C. Sreedhar
• Xinan Tang
• Kevin Theobald
• Ruppa Thulasiram
• Parimala Thulasiraman
• Xinmin Tian
• Yingchun Zhu
• J. Nelson Amaral