Program Systems Institute of the Russian Academy of Sciences
Research Center for
Multiprocessor Systems
Doctor Sergei Abramov
Supercomputers and multiprocessor
software.
 Technologies for regional computer networks
building.
 Functional programming, supercompilation
and metacomputation theories, and their
applications to practical programming.

Program Systems Institute of the Russian Academy of Sciences
The SKIF Supercomputer Project
of the Russia-Belarus Union State
Participants: structure
The standing committee
of the Union State
National Academy
of Sciences,
Belarus
UIIP of NAS of Belarus
Participants from Belarus
The Ministry of
Education and
Sciences
PSI RAS
Participants from Russia
Program Systems Institute of the Russian Academy of Sciences
2000-2004: 16 clusters produced
“К а рд и ол о ги я ”
9 /5 G
3 + 1 -1U + 4 U
In te l P -IV-1 2 6 6
“П е р ве н е ц ”
2 0 /11 G
“В М -5 1 0 0 ”
4 8 /2 6 G
“M y rin ”
8 9 /5 9 G
“К -5 0 0 ”
7 1 7 /4 1 5 G
“К -1 0 0 0 ”
2 5 3 4 /2 0 3 0 G
1 6 -3 U ;
In te l P -III-6 0 0
1 6 -2 U
In te l P -IV-1 5 0 0
8 -1 U
In te l X e o n 2 .8
6 4 -1 U
In te l X e o n 2 .8
2 8 8 -1 U ; IB 4 x; A M D
O p te ro n 2 4 8 (2 .2 )
2000
2001
2002
2003
2004
“П е р ве н е ц ”
2 0 /11 G
“С туд е н т”
11 /6 G
“Т К С ”
4 0 3 /2 3 0 G
“T-B rid g e 8 i”
4 5 /3 7 G
1 6 -3 U ;
In te l P -III-6 0 0
9 -M in iTo w e r
In te l P -III-6 0 0
3 6 -1 U
In te l X e o n 2 .8
4 -1 U ; IB 4 x
In te l Ita n iu m 2 (1 .4 )
“Ги б р и д ”
2 .4 /1 .2 G
“П е р ве н е ц -M ”
9 8 /5 7 G
“T-F o rg e 3 2 ”
11 5 /7 4 G
“T-F o rg e 4 8 ”
2 3 0 /1 8 4 G
2 -4 U + 6 U ;
In te l P -III-8 0 0
1 6 -3 U ; A M D
A th lo nM P 1 8 0 0 +
1 6 -2 U ; A M D
O p te ro n 2 2 4 (1 .8 )
2 4 -1 U ; IB 4 x; A M D
O p te ro n (2 .4 )
Н И И м ех М Г У
4 9 /2 8 G
“T-F o rg e 2 0 H B ”
8 8 /7 0 G
4 + 4 -4U + 5 U ; A M D
A th lo nM P 1 8 0 0 +
1 0 -H B ; IB 4 x; A M D
O p te ro n 2 4 8 (2 .2 )
Program Systems Institute of the Russian Academy of Sciences
Peak performance of the clusters
2001
G F lo p s
2002
2003
2004
2534
2500
717
1000
x125 ratio
98
100
48
20
10
1
2000
П е р ве нец
2001
ВМ
5100
К а рд и о- П е р ве л о ги я
н е ц -М
2002
НИИ
м ех.
М ГУ
Ги бр ид
С т удент
2003
M yrin
TF o rg e
32
С КИ Ф
К -5 0 0
Т КС Е С
1 7 1 0.0 3
2004
TF o rg e
20H B
TF o rg e
48
TB rid g e
8i
С КИ Ф
К -1 0 0 0
Program Systems Institute of the Russian Academy of Sciences
Linpack-performance of the clusters
2001
G F lo p s
2002
2003
2000
1000
472
2004
2032
x185 ratio
57
100
26
10
1
11
2000
П е р ве нец
2001
ВМ
5100
К а рд и о- П е р ве л о ги я
н е ц -М
2002
НИИ
м ех.
М ГУ
Ги бр ид
С т удент
2003
M yrin
TF o rg e
32
С КИ Ф
К -5 0 0
Т КС Е С
1 7 1 0.0 3
2004
TF o rg e
20H B
TF o rg e
48
TB rid g e
8i
С КИ Ф
К -1 0 0 0
Program Systems Institute of the Russian Academy of Sciences
2000-2004: Supercomputing Industry.
Linpack-performance Growth
Program Systems Institute of the Russian Academy of Sciences
2000-2004: Supercomputing Industry.
Linpack-performance Growth
Linpack:
performance
Four-year
growth
(times)
Linpack:
performance
Four-year
growth
(times)
TopSKIF
185
Top200
14
Top1
14
Top300
15
Top10
11
Top400
15
Top100
11
Top500
15
Program Systems Institute of the Russian Academy of Sciences
SKIF K-500
(October 2003, Minsk)
Developers
• UIIP of NAS, Belarus
• NII EVM
• The T-Platforms Company
• PSI RAS
The project was completed in
months
Program Systems Institute of the Russian Academy of Sciences
SKIF K-500: testing
(September 2003, Moscow)
The project has
been completed
in 4 months,
cost per 1 TFlops
(peak):
under $700,000
Program Systems Institute of the Russian Academy of Sciences
Minsk: К-500 (September 2003)
Peak / Linpack performance
Processors (Intel Xeon 2.8)
Number of computation nodes:
Main memory: 642 =
Disk memory: 6460 =
System network 4x4х4 3D-top
Accessory network
716,8 / 423,6
(471.6) GFlops
642 =128 items
64 items.
128 GB
3 840 GB
SCI, D336
GB Ethernet
Program Systems Institute of the Russian Academy of Sciences
Program Systems Institute of the Russian Academy of Sciences
SKIF К-1000

Developers





UIIP of NAS, Belarus
NII EVM
PSI RAS
T-platforms
Development stages




Variant investigation, concept development: December
2003 — March 2004
Specification preparation: April 2004
Preparation of tender documents, tender:
May 2004
Implementation: July 15 — October 1, 2004, 2½ months
Program Systems Institute of the Russian Academy of Sciences
SKIF К-1000
Peak performance
2,5 Tflops
 Linpack- performance
2,0 Tflops
 Efficiency factor =
80.1 %

SKIF К-1000 project cost: $ 1 800 000
(several times cheaper than the competition)
 Competitions: IBM, HP, Fujitsu-Siemens

Program Systems Institute of the Russian Academy of Sciences
SKIF К-1000: hardware
"C o re " IB ком м у та то р: 1 2 ш т у к
Уп р а вл я ю щ и й у зе л : 1 ш т у ка
"Le a f" IB ком м у та то р: 2 4 ш т у к
Р а б оч и й у зе л : 2 8 8 ш т у к
"C o re " G bE th ком м у та то р : 2 ш т у ки
"Le a f" E th ком м у та тор : 8 x 2 = 1 6 ш т у к
В с е го в 8 х 4 2 = 3 36 U
у ст а н о в л е н о 3 43 у с т р о й с т в
SKIF К-1000:
convenient components design,
easily scalable
up to 15 TFlops
Program Systems Institute of the Russian Academy of Sciences
Program Systems Institute of the Russian Academy of Sciences
SKIF К-1000
is ranked #98 in the
world (the Top500
rating)
 Only four countries
take the first 100
ranks: Union State,
USA, Japan, and China



November 2004: first place in the world according to the
TopCrunch rating (the 3 Cars Crash Test)
November 2004: first place on the territory of ex-USSR
(the Top50 national rating)
Program Systems Institute of the Russian Academy of Sciences
SKIF in the Top50 rating




On December 7, 2004 seven SKIF computers were listed
of fifty most powerful supercomputers of the
Commonwealth of Independent States (Тор50,
http://www.supercomputers.ru) and took the following
places: 1, 6, 8, 20, 22, 32, 34
Installations of the SKIF family take 14% of all places in
the list
Their total peak performance amounts to 25% from the
sum of all peak performances of all the systems presented
in the Top50 list
Their total Linpack performance amounts to 33% from the
sum of all Linpack performances of all systems presented in
the Top50 list
Program Systems Institute of the Russian Academy of Sciences
Cluster-level Software Suite
Linux-SKIF OS kernel
 PVFS-SKIF — parallel file
system
 OpenPBS-SKIF — batch
queue
 FLAME-SKIF — the system of
monitoring and control (reset,
power on/off) over installations
of the SKIF family

Program Systems Institute of the Russian Academy of Sciences
Cluster-level Software Suite

OpenTS — open architecture
T-system
 the TG++ compiler for the T++
language
 the TF2TC
translator (T-Fortran  T++)

TDB — distributed interactive
debugger for MPI programs
supporting T-programs
debugging (TotalView
counterpart)
Program Systems Institute of the Russian Academy of Sciences
Cluster-level Software Suite
6 software applications in the
OpenTS environment
 12 adapted free packages,
libraries, and applications
 14 in-house designed
applications (3 of them in the AI
field)
 The platform is compatible with
commercial engineering
packages (over 6 fields of
application)

Program Systems Institute of the Russian Academy of Sciences
ServNET:
cluster management (June 2003)
RS-485 links :
up to 125 nodes, up to 1,000 meters
Several links (equaling the number
of COM-ports) may be connected to
the control station.
Program Systems Institute of the Russian Academy of Sciences
ServNET: simple hardware





Single-sided board.
One Atmel microcontroller
Two Maxim chips (MAX232A и MAX487)
One chip for ferric RAM (FRAM)
RAMTRON.
.
Few auxiliary components
Program Systems Institute of the Russian Academy of Sciences
ServNET: functionality
Selective node reset (SW: groups).
 Selective switching on/off of node’s power
supply (SW: groups, also “smooth”, with a
target latency “between nodes”).
 Access to the serial console of the node
allowing...

Program Systems Institute of the Russian Academy of Sciences
Access to the serial console of the node allows the
following:





To change the parameters of the BIOS node.
LILO: select an OS to be loaded and the loading
parameters of the Linux core.
To execute any commands in a console regime.
To monitor critical messages of the OS.
To read several last console messages from the
nonvolatile memory of the ServNET board after
failure.
Program Systems Institute of the Russian Academy of Sciences
ServNET: production
Development – PSI of RAS,
Production – Computer Research Institute,
Minsk.
 A pilot lot of more than 200 items has been
produced. ServNET has been installed in:






T-Forge32 (16 nodes).
“SKIF К-500” (64 nodes) and other clusters in Minsk.
Clusters in Pereslavl.
Clusters in the Research Computing Center of the
Moscow State University (58 ServNET boards).
ServNET v.2: 2004, developed by PSI of RAS
and T-platforms, produced by T-platforms.
Program Systems Institute of the Russian Academy of Sciences
SKIF-ServNet v.2

The size has become half as large (66 × 33 mm) if
compared with the previous version.
Program Systems Institute of the Russian Academy of Sciences
Design and Development of
Regional Telecommunication Systems
Head: Yury Shevchuk, Ph.D.
 The
Botik Lab.: Activities
Develops cost effective solutions for regional
computer networks (BOTIK Technologies)
Implements the technology in Pereslavl region
for testing and enhancement
Transfers the approved technology to Russian
regions and the CIS countries
Program Systems Institute of the Russian Academy of Sciences
The BOTIK Network
 Started
in 1994: 11-year history
 State-of-the-art
More than 900 subscribers
More than 3000 networked PCs
External traffic: 250 Gb/month
 4 technology transfer contracts
Almaty, Kazakhstan
Moscow
Ramenskoye, Moscow region
Severomorsk, Murmansk region
Program Systems Institute of the Russian Academy of Sciences
The BOTIK Technologies: Core
 PC
router
IBM PC + Linux + in-house designed HW & SW
 Other
hardware solutions
Resistance to low-quality power supply,
temperature difference, and vandalism
Self-recovery and self-monitoring of devices
Wireless solutions for rural regions
Etherbox: sensor networks device
 Software
solutions for regional networks
Network Administration System: +10,000 lines…
Program Systems Institute of the Russian Academy of Sciences
Urban Networks: Pereslavl-Zalessky
Program Systems Institute of the Russian Academy of Sciences
Rural Networks: Pereslavl Region
Program Systems Institute of the Russian Academy of Sciences
Functional programming,
supercompilation and
metacomputation theories, and
their applications to practical
programming
Program Systems Institute of the Russian Academy of Sciences
Functional programming
in IPS RAS
Supercompilation, program specialization.
 Metacomputation, semantics modifiers – reusing
“semantics components”.
 Implementation of functional programming
languages.
 Applied computer algebraic libraries.

Program Systems Institute of the Russian Academy of Sciences
Reasons for program optimization

The methods can do almost nothing with programs that are
already carefully optimized by a professional programmer in
a lower-level language. The methods can clean a program of
natural inefficiencies if the program has been developed
“inefficiently” in a structured way, using various “high-level”
techniques like
1.
interpretation of specialized application-oriented languages
2.
component programming from libraries of general re-usable
and
software, etc.
Thus, the methods are directed to provide degrees of
freedom for new software technologies rather than to
optimizing programs written in an old style.
Program Systems Institute of the Russian Academy of Sciences
Specialization: the main idea
Let human be a program with two parameters
knowledge and problem. Then creating a specialist
humanknowledge from human and knowledge is a good
example of specialization:
humanknowledge(problem) = human(knowledge,problem)
Specialist humanknowledge can solve problems much
quicker than an ordinary human when the problems
are covered by his specific knowledge.
Program Systems Institute of the Russian Academy of Sciences
A number of tasks for specialization
<F x0, y>
The first argument of a program F is given, while the second is unknown.
<F <G x, y> , z >
Let two programs F and G be given, specialize a composition of
applications of the programs. That is specialization with respect to a
context of application.
<IntL ( Program )
e.data
>
<GO
>
An interpreter IntL of a programming language L is specialized with
respect to a given program. IntL is written in a language M, while the
program is written in L, so we expect an optimal program written in M as
a result of specialization. Thus a specializer may be used as a compiler
from L into M, where M is the subject language of the specializer.
Program Systems Institute of the Russian Academy of Sciences
What is supercompilation ?
Supercompilation is a technique of specialization of programs written in
a functional programming language. The technique was introduced in the
1970s by V. F. Turchin. He proposed a task of creating tools to observe
operational semantics of a program, when a function F that is to be
computed by the program is fixed. As a result of such observations a new
algorithmic definition of an extension of the function F must be
constructed. His ideas were studied by a number of authors for a long
time.

The main aim of a supercompiler is to perform as many actions of a given
parameterized application of a program uniformly on the parameters as
possible.


Also supercompilers can be used
• as theorem provers for program verification;
• as compilers by specialization of operational specifications;
• for porting a non-standard semantics from one programming
language to another; by specialization of a semantics modifier.
Program Systems Institute of the Russian Academy of Sciences
The Supercompiler SCP4

is an experimental specializer for a functional
language Refal-5. (There are no special restrictions
on the input language.) SCP4 has been
implemented once again using Refal-5. Sources of
the supercompiler, executable modules and sources
of Refal-5 are available for immediate free
download:
http://www.botik.ru/pub/local/scp/refal5/
Windows 98
Windows NT/2000/XP
Linux (Intel)
Program Systems Institute of the Russian Academy of Sciences
Specialization of interpreters
<IntL ( Program )
e.data
<GO
>
>
An interpreter IntL of a programming language L is
specialized with respect to a given program. IntL is written in
a language M, while the program is written in L, so we expect
an optimal program written in M as a result of specialization.
Thus a specializer may be used as a compiler from L into
M, where M is the subject language of the specializer.
Semantics modifiers (a class of programs that allow the
development of general and reusable “semantics
components”) can be specialized alike interpreters.
Program Systems Institute of the Russian Academy of Sciences
Verification of parameterized systems
by the supercompiler SCP4
 Successful
experiments on verification of
cache coherence protocols:
IEEE Futurebus+, MOESI, MESI, MSI, “Illinois”,
“Firefly”, “Berkeley”.
 More
parameterized protocols:
Java Meta-Locking Algorithm, Reader-Writer
protocol.
Program Systems Institute of the Russian Academy of Sciences
References
[1] Abramov S.M., and Glueck R. From standard to non-standard semantics by semantics
modifiers. International Journal of Foundation of Computer Science, Vol. 12 No. 2,
pp:171-211, 2001.
[2] Nemytykh A.P., and Turchin V.F. The Supercompiler Scp4: sources, on-line
demonstration. http://www.botik.ru/pub/local/scp/refal5/ ,2000.
[3] Nemytykh A.P., The Supercompiler Scp4: General Structure., LNCS vol. 2890,
pp.162-170, 2003.
[4] Nemytykh A.P., A Note on Elimination of Simplest Recursions. In Proceedings of
the ACM SIGPLAN Asia-PEPM'02, 138-146. ACM Press, 2002.
[5] Korlyukov A.V., and Nemytykh A.P., Supercompilation of Double Interpretation.
(How One Hour of the Machine's Time Can Be Turned to One Second). (In English),
Vestnik natcional’nogo tekhnicheskogo universiteta “Khar’kovskogo politekhnicheskogo
instituta”, Khar’kov, No. 1, 2004.
[6] Lisitsa A., and Nemytykh A.P., Verification via Supercompilation.
http://www.csc.liv.ac.uk/~alexei/VeriSuper/ , 2005.
Descargar

Исследовательский центр мультипроцессорных систем …