Performability and Grid in an On Demand World

1
Performability and Gridin an On Demand World

• Robert Berry,
• IBM Corporation

2
Good morning and thank you
3
Outline

• Performance and Reliability are important
• The Grid is an important component of business
  computing, today, and for the future
• On Demand the drive for ultimate flexibility
  and integration
• Leads us to Grid
• The Grid in business
• Typical Examples of commercial Grid applications
• A different kind of example
• Trends, Consequences, Questions for
  Performability

4
Performance, Reliability are important

• Reliability, Quality
• A growing part of software development budget
  is spent on performance, test and service
• Problem rates on new technologies (e.g., J2EE)
  are high and climbing
• New solutions are required.

5
On Demand Business The Why
An on demand business is an enterprise whose
business processesintegrated end-to-end across
the company and with key partners, suppliers and
customerscan respond with speed to any customer
demand, market opportunity or external threat.
6
Present Reality Functional Automation

• Limitations of Tight Vertical Integration
• Monolithic applications cant be reused
  economically, efficiently
• Ad hoc integration creates connections that are
  difficult to change/maintain
• Lack of standards limits ability to deliver
  meaningful interoperability

7
What Business needs Horizontal Process
Integration

• Advances that make it possible
• Standards for creating services and enabling them
  to communicate are agreed upon by major vendors
• Infrastructure that supports self-defined,
  loosely coupled services has emerged
• Tools to incorporate existing assets are
  available
• Automation and virtualization of systems
  resources readily available

8
What Business needs Horizontal Process
Integration

• Increased focus on business flexibility
• Relentless use of open standards to enable
  communication and integration across the value
  net
• Partner to sharpen focus and respond to
  opportunities and threats
• Requires an on demand IT environment
• Ability to repurpose application functionality in
  order to support horizontal processes
• Automation and virtualization of resources

9
IBMs On Demand model is a fundamental shift in
the business model and technology
Static
On Demand
Business process decision-making
Collaborative, integrated value nets
Rigid organizational structure
Dynamic, adaptive, learning
Slow and steady economic growth
Unpredictable fluctuations
Business
Long-term product lifecycles
Shortening product lifecycles
Passive operational risk management
Proactive risk management increased focus on
privacy and security
Fixed costs
Variable costs
Proprietary systems
Open, integrated systems
Technology
Labor-intensive
Self-healing, self-managing systems
Users adapt to technology
Technology adapts to users
10
On Demand Operating Environment
Focus on Virtualization and Grids
Integration People Process Information
Anywhere, any time, from any device
Transactional Processes
Information Management
Collaboration
Application Development, Deployment Maintenance
Open Standards-based
Business Objectives and Policies
Business Objectives and Policies
Virtualization Engine
Grids
Distributed Systems
Storage
Servers
Network
11
Grid is it commercially real?

• Several commercial examples
• Charles Schwab
• Hewitt Associates
• A somewhat different example
• Games

12
Charles Schwab
Business Analytics
Challenge

• Reduce the processing time on an existing
  wealth-management application to improve customer
  service.

Technology Benefits
Solution

• IBM
• Linux
• Globus Toolkit
• IBM Infrastructure Technology Services
• IBM Research

• Technology Benefits
• Reduced processing time from four minutes to 15
  seconds
• Leverages existing infrastructure
• Grid-enabling many more applications

• Business Benefits
• Increased customer satisfaction by responding to
  inquiries in real time
• Enabling Schwab to move from a low-cost
  transactional broker to an advice-based wealth
  manager

We believe that Grid computing has the
potential to greatly improve our quality of
service and be a truly disruptive technology.
Oren Leiman, Managing Director, Charles Schwab
13
Hewitt Associates LLC
Business Analytics

• Challenge
• Create Grid Computing environment to
• Contain expenses for CalcEngine valuations
• Maintain or improve availability, response time
  scalability
• Insure personal-data security
• Capitalize on existing application code
• Cooperate with z/OS Sysplex CICS Calling
  Environment
• Enable smooth and orderly migration to change
• Solution
• Grid Computing environment includes
• IBM zSeries server
• IBM eServer BladeCenterTM servers
• Linux Red Hat v8.0
• Business Partner DataSynapse GridServer

• Benefits
• Efficiently uses of the combined processing power
  of their heterogeneous environment
• Experienced an immediate 10 faster response time
  with the first application deployment
• Open architecture enables Hewitt to easily deploy
  additional applications
• Increased processing speed reduced cost per
  transaction
• Reduced operational costs improves
  competitiveness in their industry segment

14
HR Outsourcer Business Background
zSeries Model 900

• zSeries Sysplex
• 16000 Mips
• 71 CPs
• 4 CPs for CalcEngine
• More CPs for development

CICS
AAA COBOL
BA0111 COBOL
CaIcEngine Smalltalk

• Real time policy calculation
• Valuate participant scenario with respect to
  employer constructs
• Policies defined by corporate plans and
  governmental rules
• Custom interpretive language worksheet creation
• Valuations are an expensive business
• COBOL CICS applications serialize person data
  through VSAM
• Smalltalk Operations can take 1-50 seconds of
  zSeries computing
• Arrival rate is erratic, causing divergent
  response time
• Grid cost is justified by the flexibility for new
  deployment

15
Integrating Smalltalk Into A Grid
Grid Servers (xSeries)

• IBM VisualAge Smalltalk Connectors
• Support for APPC, CICS, CPI-C, MQI and TCP/IP
  (sockets and RPC)
• Version 6.0.1 became a pre-req
• Required DB2 access from Engines
• Prototype Selection
• Tried OGSA v3 in February, but
• Keep it simple!
• Use Java socket code
• Provide enough middle points to be able to
  watch/debug request flow

Binding App.
Linux
Linux
VMWare or Blades
16
OptimalGrid

• A Research project at IBM Almaden Researcher
  Center
• Makes developing and running grid application
  easy by hiding the complexity of using the grid
• A different approach from the API or Toolkit
  application development environments no need to
  educate developers on complicated grid
  technologies
• Application code, Management Infrastructure, and
  Manager written in Java so it runs anywhere Java
  runs
• Non-Java application integration possible via JNI
  and MathShell

17
What if.?

• You could create a NEW GENRE of massive player
  online game
• Where you can FLY through a huge game world
• Where you can SEE objects in the other server
• Where you can INTERACT with the entities on the
  other servers

Server 1
Interacting with Entities in other servers
Server 4
Range of sight For player 1
Server 2
Range of sight For player 2
Server 7
Server 3
Server 8
Server 6
18
What if.?

• You could create a MASSIVE multi-player online
  game
• Where the number of players are NOT LIMITED by
  the server hardware
• Where everybody is in ONE WORLD, not several
  different shards
• Where massive battles with over 10,000 PLAYERS
  are possible

ONE Gridified Game World
Three separate Parallel Game Worlds
3 Game world server (MAX 3000 players) Outage
due to sudden increase in players
1 Game world server ( 2000 players )
Hello Player ABC
Hello player XYZ
2 Game world server ( 1000 players )
Game environment not limited by server hardware
Come join the big battle!
19
What if.?

• Your game servers could load balance the game
  DYNAMICALLY
• Where the servers will allocate resources
  AUTOMATICALLY
• Where the servers will partition the game for
  OPTIMAL PERFORMANCE

Game map zone Blue
Game map zone Red
Game map zone Yellow
Huge User Movement From Map Zone BLUE To Map
Zone RED
Grid Resource Dynamically Re-allocated
with Optimal configuration
20
Massively Multiplayer Online Game (mmog)
Demonstration Based on Quake2

• 870 unique total client code downloads
• 67 IBM domains worldwide
• 18 Linux clients 82 windows

21
OptimalGrids Autonomic feature hides the
complexity of Grid from the Application
Developer, while resolving pain points

• For Game Developers, Hosts, and Aggregators of
  MMOG
• OptimalGrid can
• Automatically partitions and distributes
  any game world
• Seamlessly moves players and objects between
  servers
• Game migration is easy
• Original game design is
• preserved

22
Performability in an on demand world
Performance
Reliability
23
Flexibility and Responsiveness drive On Demand
And they, in turn, drive Virtualisation (Grid)
And also have consequences for Performability
Componentisation
Flexibility
Finer modularity
Reuse
Simplification
Responsiveness
Virtualisation at all levels of Resources
Open Stds
Autonomic control
Security
Performance
Reliability (predictability)
24
Trends, their Consequences, and Questions

• Security
• A classic antithesis for performance
• A critical and growing requirement for business
• But a growing problem we all face, e.g.,
• Viruses
• 6.5M in 2003
• Antivirus scans - 3 hours of downtime
• Overhead for J2EE security
• managers ---------------------------------?
• This will continue to grow
• 6M Viruses so far in 2004

25
Trends, their Consequences, and Questions

• Security

• Typically drives performance down
• - pathlength goes up
• - scalability is constrained

Performance
Security
Reliability

• Typically enhances reliability
• Availability also goes down with
• virus scans, etc.

Open, Grid environments scale up security issues,
and mandate more attention. Do we need to
model these interactions explicitly? Are
present techniques adequate? Appropriate? e.g.,
to capture viral spreading patterns, containment
26
Trends, their Consequences, and Questions

• Scalability
• Whatever we think is large and complex now, wont
  be in the future.
• We see this in some of the grid examples we are
  familiar with
• Games motivates another jump in scale
• 10,000 users at a time
• Communities of 500,000s of users
• Wireless gaming will drive server demand, and
  strain adaptive capabilities even harder
• Security issues will multiply profoundly
• Gaming is real commerce
• People buy virtual things e.g., castles, etc.,
  on ebay!

27
Trends, their Consequences, and Questions

• Complexity
• A growing concern for software in general
• New technologies/disciplines e.g.,
  componentisation, Aspect Oriented Software
  Development - must help, but for now, we are
  faced with

• One factor that contributes to growing security
  overhead
• Also to resulting models of reliability and
  overall fragility

28
Trends, their Consequences, and Questions

• Componentisation and Reuse
• Demands modularity and reuse in models
• Demands incremental, scalable, composable models
• Autonomic Systems
• What is a failure in a self-healing system?
• What are the reliability characteristics of a
  system that is adaptive?
• How is this to be characterised, measured,
  modelled?

29
Trends, their Consequences, and Questions

• Autonomic Systems
• What is a failure in a self-healing system?
• What are the reliability characteristics of a
  system that is adaptive?
• How is this to be characterised, measured,
  modelled?

30
Trends, their Consequences, and Questions

• Business focus is needed
• Growing Recognition of the importance to IT of
  Business Events, Business Processes
• For deeply technical people this is a hard
  concept
• But, technology exists to solve business
  problems
• There are some interesting opportunities for
  performability as well. E.g.,
• Reliability as a factor in service level
  objectives, agreements
• Dynamic, operational performability
• Measuring and Characterising Business
  Performance Business Reliability. The
  Reliability of Services.
• Business Performance Management monitors and
  visualizes the behaviour of business processes by
  correlating IT events and business activities,
  thus enabling performance optimization according
  to business goals

31
Summary

• Performance, Reliability are growing business
  concerns
• The Grid solves real business problems and is
  part of the overall solution in delivering
  increasing flexibility and responsiveness.
• Dependency on Software is growing in all levels
  of IT solutions
• Complexity is growing
• Security is a growing requirement, and the
  tradeoffs between security and performance need
  to be better understood
• The IT focus is moving increasingly higher up the
  software stack indeed, beyond the software into
  the business domain.
• Reliability, Performance and Availability are
  more complex with dynamic provisioning and
  autonomic control

32
Thank You!

• Questions?

33
Getting OptimalGrid

• Available for download on Alphaworks
• www.alphaworks.ibm.com/tech/optimalgrid
• Project pages
• www.almaden.ibm.com/software/ds/OptimalGrid
• E-week
• http//www.eweek.com/article2/0,3959,1227299,00.as
  p
• Slashdot
• http//games.slashdot.org/article.pl?sid03/08/22/
  0351229modethreadtid112tid127tid136tid18
  6tid187tid206)
• MIT Technology Review
• Researchers
• James Kaufman kaufman_at_almaden.ibm.com
• Toby Lehman toby_at_almaden.ibm.com
• Glenn Deen glenn_at_almaden.ibm.com

Text slide withfilm strip images
34
Trademarks

• Hewitt is a trademark of Hewitt Associates, LLC
• Charles Schwab is a trademark of The Charles
  Schwab Corporation.
• Java and all Java-based trademarks are trademarks
  of Sun Microsystems, Inc. in the United States,
  other countries, or both.
• Microsoft, Windows, Windows NT, BizTalk, and the
  Windows logo are trademarks of Microsoft
  Corporation in the United States, other
  countries, or both.
• Gartner is a registered trademark of Gartner,
  inc., or its Affiliates
• Solaris is a trademark of Sun Microsystems, Inc.
  in the United States, other countries, or both.
• UNIX is a registered trademark of The Open Group
  in the United States and other countries.
• Intel, Pentium, Pentium Pro, Pentium II, Pentium
  III are trademarks or registered trademarks of
  Intel Corporation or its subsidiaries in the US
  and other countries.
• HP-UX is a registered trademark of Hewlett
  Packard Company.
• Linux is a registered trademark of William R.
  Della Croce, Jr. (last listed previous owner was
  Linus Torvalds)