Transcript Presentation

Silicon Graphics, Inc.
SAN over WAN - a new way
of solving the GRID data
access bottleneck
Presented by:
Dr. Wolfgang Mertz
Business Development Manager for
Storage in EMEA
[email protected]
Cracow ‘03 Grid Workshop
Data Growth Trends
(in Terabytes)
7,000,000
6,000,000
5,000,000
4,000,000
3,000,000
2,000,000
1,000,000
-
From 2001 to 2005
it is projected to grow
at 83% CAGR
From 1998 to 2000
Storage Shipped grew
at 78% CAGR
1998
1999
2000
2001
2002
2003
2004
2005
Data under management in an HPC environment is currently growing at over 100%/year.
Source: Lyman, Peter and Hal R. Varian, "How Much Information", 2000. Retrieved from
http://www.sims.berkeley.edu/how-much-info on 12/19/2002.
Cracow ‘03 Grid Workshop
Page 2
2 Buzzwords in IT Industry
• Server Consolidation
– maybe in a commercial environment
– usually not in a technical environment
• a hammer is a hammer, a screwdriver is a screwdriver
• an HPC system cannot be used as a HPV system
• Storage Consolidation
– DAS -> NAS -> SAN
Cracow ‘03 Grid Workshop
Page 3
History of Storage Architectures
DAS - Direct Attached Storage
• pro
–appropriate performance
• con
–distributed, expensive administration
–data may not be where it is needed
–multiple copies of data stored
Cracow ‘03 Grid Workshop
Page 4
History of Storage Architectures
NAS - Network Attached Storage
• pro
–centralized, less expensive administration
–one copy of data
–access from every system
• con
–network performance is the bottleneck
Cracow ‘03 Grid Workshop
Page 5
History of Storage Architectures
SAN - Storage Area Network
• pro
–centralized administration
–performance equivalent to DAS
• con
–NO FILE SHARING
–multiple copies of data stored
Switch
Cracow ‘03 Grid Workshop
Page 6
How does that translate to a GRID Environment?
• Storage Consolidation
– useful in a local environment (GRID node)
– does not work between remote GRID nodes
• Current Data Access between GRID Nodes
– Data has to be copied before/after the execution of a job
– Problems
• copy process has to be done manually or included in the job script
• copy can take long
• multiple copies of data
– additional disk space needed
– revision problem
Cracow ‘03 Grid Workshop
Page 7
What if...
• ... a SAN would have the same file sharing capability as
a NAS?
• ... one could build a SAN between different
buildings/sites/cities and not loose performance?
Cracow ‘03 Grid Workshop
Page 8
Storage Area Networks (SAN)
The High Performance Solution
A first step:
• each host owns a
dedicated volume
consolidated on a
RAID array.
•Storage
management is
centralized.
LAN
SAN
•Offers a certain
level of flexibility.
Cracow ‘03 Grid Workshop
Page 9
SGI InfiniteStorage Shared FileSystem (CXFS)
A unique high
performances
solution:
•Each host shares
one or more volumes
consolidated in one
or more RAID arrays.
LAN
•Centralized storage
management
•High modularity
•True High
Performances Data
sharing
SAN
•Heterogeneous
Environment
Cracow ‘03 Grid Workshop
Page 10
Fibre Channel over SONET/SDH
The High Efficiency, Long Distance Alternative
Hours
250
Hours to Send 1 TeraByte
200
OC-12 IP
OC-12 SONET/SDH
150
100
50
Distance
(kilometers)
0
0
1000
New Boston
York
Cracow ‘03 Grid Workshop
2000
3000
Chicago
4000
Denver
Page 11
Data re-transmission
due to IP packet loss
limits actual IP
throughput over
distance
LightSand Solution for building a Global-SAN
WAN
Client
Servers
LAN
LAN
IP Router
IP Router
Client
Servers
DWDM
SAN
SAN
Dedicated
Fiber
Fibre Channel
Switch
IP
Storage
Tape System
Cracow ‘03 Grid Workshop
Fibre Channel
Switch
SDH
SONET
Tape System
FC
SONET
Page 12
Storage
LightSand Products
• S-600
– 2 ports FC and/or IP 1Gb/s
– Point-to-point SAN interconnect over SONET/SDH OC-12c (622 Mb/s
bandwidth)
– Low latency (approximately 50 µSec)
• S-2500
– 3 ports FC and/or IP 1Gb/s
– Point-to-point SAN interconnect over SONET/SDH OC-48c (2.5 Gb/s
bandwidth)
– Point-to-multipoint SAN interconnect over SONET/SDH (up to 5 SAN islands.
622 Mb/s per link)
– Low latency (approximately 50 µSec)
Cracow ‘03 Grid Workshop
Page 13
Data Movement Today –
A Recent Case Study
IP Network
Server
Sandia
National
Laboratory
(SNL)
Los Alamos
National
Laboratory
(LANL)
Fibre Channel
Storage Area
Network
Server
Scientists at LANL currently dump 100GB of supercomputing data
to tape and FedEx it to SNL because it is faster than trying to use
the existing 155Mb/s IP WAN connection
–
Actual measured throughput of 16Mb/s! (10% bandwidth utilization)
http://www-unix.mcs.anl.gov/discovery/wufeng.htm
Cracow ‘03 Grid Workshop
Page 14
Fibre Channel
Storage Area
Network
The Better Way – Directly Between Storage
Systems
IP
Network
Server
FC SAN
Server
Local
Data Center
LightSand
Gateway
Remote
Data Center
Telco
SONET/SDH
Infrastructure
FC SAN
LightSand
Gateway
Using LightSand gateways, the same data could be
transferred in a few minutes!
Cracow ‘03 Grid Workshop
Page 15
What does that mean for a GRID Environment?
• Full Bandwidth Data Access across the GRID
• No Multiple Copies of Data
– avoid the revision problem
– do not waste disk space
• Make GRID Computing more efficient
GDAŃSK
GDAŃSK
POZNAŃ
POZNAŃ
WROCŁAW
WROCŁAW
WARSZAWA
ŁÓDŹ
ŁÓDŹ
KRAKÓW
KRAKÓW
Cracow ‘03 Grid Workshop
Page 16
Highly Integrated,
Massively Scalable Systems
HighPerformance
Computing
Advanced
Graphics
Cracow ‘03 Grid Workshop
Storage
Page 17
SGI InfiniteStorage Product Line
DAS
NAS
High Availability
Redundant Hardware and FailSafe™
XVM
SAN
Legato NetWorker,
XFS™ Dump, OpenVault™
Data Sharing
HSM
Data Protection
High Availability
Data Protection
HSM
SGI Data Migration Facility (DMF),
TMF, OpenVault™
Data Sharing
XFS, CIFS/NFS, Samba,
ClusteredXFS (CXFS™),
SAN over WAN
Storage Hardware
TP900, TP9100, TP9300, TP9400, TP9500,
HDS 99x0,
STK Tape Libraries, ADIC Libraries,
Brocade Switches,
NAS 2000, SAN 2000, SAN 3000
Cracow ‘03 Grid Workshop
Choose only the integrated
capabilities you need
Page 18
www.sgi.com/products/storage