Nordic DataGRID Facility

1
Nordic DataGRID Facility

• Status

2
Mega-Science

• The coming era in science is expected to be
  dominated by Mega-Science
• 104 scientists on a project
• Huge data productions
• Collaborations within the projects
• Examples
• CERN LHC
• ALMA

3
NDGF
4
Who?

• Denmark
• Finland
• Norway
• Sweden

5
Purpose
The NDGF main focus is to create and operate the
mechanisms, organizational structures, and
development projects needed to leverage the
national investments in computing infrastructures
and meet the long term vision. It is important to
underline that the suggested Nordic commitment
does not substitute relevant and needed national
funding for competence and infrastructure
build-up in the participating Nordic countries,
rather NDGF will facilitate sharing of these
national resources creating added Nordic value.
6
What?
7
What?
8
What?
Certificate Data Version 3 (0x2)
Serial Number 2 (0x2) Signature
Algorithm md5WithRSAEncryption Issuer
CDK, STDenmark, OIMADA, OUSDU,
CNNorduGridCA Validity Not
Before Oct 8 120256 2004 GMT Not
After Oct 8 120256 2005 GMT Subject
CDK, STDenmark, LIMADA, OSDU, CNBrian
Vinter Subject Public Key Info
Public Key Algorithm rsaEncryption
RSA Public Key (1024 bit)
Modulus (1024 bit)
00d88649c53d5a0532c443ec786488
snip Exponent
65537 (0x10001) X509v3 extensions
X509v3 Basic Constraints
CAFALSE Netscape Comment
OpenSSL Generated Certificate
X509v3 Subject Key Identifier
AC12726F0BEC4D386B3EC2304610FDC22A
E1D1D6 X509v3 Authority Key
Identifier keyidFA748F4C1D
2B9A14AC93D74155E301BC4CED2B87
DirName/CDK/STDenmark/OIMADA/OUSDU
/CNNorduGridCA serial00
Signature Algorithm md5WithRSAEncryption
1e47530f3f75e0cee7c0188f94a00d39d6
37 e9bd96aa437c7ba1db4a54cd3
682ff2d71cb 91dbfcb2f21f0f37
a4b50c6bcf554d23cca3
f1f0f99a4e572626c7de0f494d283a6bbb
00 6b16c4e310efb7b8c96d1ad16
3d46d58d69d 9f3f47aab33b3b3f
b47938c69d53675180ce
c52c2307b5394715b3b9255e9b4044bb29
31 2193
9
GRID
10
The GRiD

• Named after the power-grid
• Sometimes referred to as the information power
  grid
• Like the power-grid GRID should be powered by
  large installations
• not individual generators

11
Philosophy

• Current Internet only allows access to
  information
• The GRiD should provide access to any desired
  resource
• CPU/SuperComputers
• Storage
• Applications

12
Why is GRID possible now

• We now have the security in place
• Network performance is now here

13
Transparent Remote File Access

• Huge input files incur a number of problems
• Download time vs. total execution time
• Job execution on the resource is delayed
• Storage requirements on resources
• Often only small scattered fragments of input
  files are needed
• How about automatic on-demand download of needed
  data?

14
Experiments

• 4 experiments
• Overhead read a one byte file
• I/O intensive application Checksum a 1 GB file
• I/O balanced application Process a 1 GB file
• Partial file traversal Search a 360 MB B tree
  for a random key
• 3 test setups
• Local execution
• Copy model
• Remote access model

15
Baseline Performance
16
Latency tests
17
Checksum
18
Balanced
19
B Tree
20
Examples
21
Use resources better
Dalton Portal
22
Chemistry / Dalton

• Dalton is a very commonly used application within
  chemistry
• Large number of functions to support quantum
  chemistry
• Some models runs excellent on cluster-computers
  while others require large shared memory
• Grid can help us

23
Dalton Cluster or SMP? Today
24
Dalton Cluster or SMP? Today
25
Dalton Cluster or SMP? Tomorrow
GRID
26
Dalton Cluster or SMP? Tomorrow
GRID
27
Dalton Cluster or SMP? Tomorrow
GRID
28
Dalton
29
Dalton Portal
30
Dalton
31
Use more resources
GRID BLAST
32
Sequence alignment the challenge

• Data volume doubles every 6-12 months
• CPU power doubles every 18 months (Moore's law)
• Needed CPU power for a BLAST-search is linearly
  related to the data volume

This slide is borrowed from Jan Host Jensen, Novo
Nordic
33
GRID BLAST

• First prototype with BLAST compatible interface
  ready
• Outstanding problems include
• A naming convention for different versions of the
  databases
• Security many users are unwilling to send their
  own databases to the Grid

34
Bio-BLAST

• Grid BLAST
• Split the databases in blocks
• Match block to block
• Assemble the results

35
Bio-BLASTE.Coli vs. Human
36
Bio-BLASTE.Coli vs. Human
37
Bio-BLASTE.Coli vs. Human
38
Bio-BLASTE.Coli vs. Human
1GB
39
Bio-BLASTE.Coli vs. Human
1GB
512 MB
40
Use resources easier
GRID VCR
41
GRID VCR

• A GRID enabled VCR
• Allows programming as a GRID job
• Allows sharing of recordings in a virtual
  organization
• Really just a prototype for any apparatus you
  might want to share
• Or just ease access to

42
Grid VCR
(executable"VCR_HOME/vcrrecord.sh")
(arguments-t 000010) (runtimeenvironment"VCRR
ECORD/DK/TV2") (jobName"ng-vcrrecord-job")
(outputFiles("video/.avi" "")("video/" ""))
(stdout"stdout") (stderr"stderr")
(CPUTime100) (startTime"2004-09-28 0852")
PLAY MOVIE
43
Own less resources
Grid terminal
44
The Grid Terminal

• If all applications are best run on the Grid
  anyway why not just do it?
• The Grid terminal is a 500MHz VIA processor based
  machine (clocked down to 250MHz)
• Used 15W
• Has no movable parts so no noise
• All applications are run on Grid

45
The Grid Terminal
46
Grid terminal
47
Use resources easier
High Bandwidth
48
High Throughput application Experiments
49
High Throughput application Experiments

• The experiment was hooked up with the ATLAS
  data-challenge 2
• A cluster was set up in Odense to simulate an
  accelerator
• Work-clusters were placed in
• DIKU 1Gb link
• NBI 1 Gb link
• IMM 1 Gb link
• IFD Forskningsnettet
• Sweden - Nordunet
• Norway - Nordunet
• Canada Internet ?

50
The Setup
51
Conclusions

• Using Grid allowed the setup to run continually
  though some sites crashed or disappeared during
  the test
• The 10Gb switch could not switch at 10Gb
• Thus we ended up with 1Gb VLan links in the 10Gb
  fiber
• We were able to saturate the network
• Not reaching 10Gb but managing to stall both
  producers and consumers

52
Towards an Open Compute Market
53
PLans
Within the NDGF core facility are several
responsibilities the management of a shared
production Grid, integrated middleware
development, an empowered committee to decide on
cross national policy issues, user and
application maintenance and possibly a Nordic
Grid Forum to coordinate academic and industrial
issues and initiatives.
54
Towards an Open Compute Market

• We move towards an economy based open Nordic
  compute market
• Users will have GRID accounts
• but no longer machine accounts
• Jobs are submitted to GRID where the centers
  compete to win the task

55
NDGF Bourse
56
One Motivation
57
Advantages

• Competition will spread
• between centers
• to vendors
• Larger market for niche architectures
• Clusters are currently flavor of the day
• SMPs are kept quite small
• No room for
• Vectors
• Reconfigurable architectures

58
Organization

• NDGF will provide a common Nordic market for
  computing
• Shared access to all Nordic resources
• Common storage
• Strong self supported middleware
• NDGF should focus on science not GRID!

59
NDGF Tasks
60
Evaluation
The Evaluation Panel agrees with the idea of
establishing a common Nordic grid. The Nordic
countries, individually, appear to have
difficulties with establishing the sufficient
critical mass to become attractive collaboration
partners in the larger grid projects in Europe
and the rest of the world. But the establishment
of NDGF will ensure for scientist in the
Nordic countries a grid infrastructure which
is of sufficient critical mass to join various
scientific projects needing grid and high
performance computing in Europe and the rest of
the world.
61
(No Transcript)