Java程序辅导

Thomas J. Watson Research Center
PO Box 218
Yorktown Heights, NY  10598
 Challenges and Opportunities in 
Autonomic Computing
June 25, 2002
presentation to ICS'02
Alfred Z. Spector
VP, Services & Software
IBM Research
aspector@us.ibm.com
Copyright IBM 2002
1
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Abstract
Significant advances are required to make systems more adaptive 
to the growing range of impulses affecting them and to reduce 
their total cost of management. Progress seems to require 
significant innovation in adaptive techniques, systems architecture, 
software engineering, and standards. In this presentation, I will 
survey the space of the requirements and draw example problems 
from real systems. I'll then discuss the space of our research at 
IBM and highlight some of the more compelling research projects 
we are doing in the area. I'll conclude with a summary of some key 
challenges for the broader community as they relate to autonomic 
computing.
2
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Introduction
Autonomic Computing
Motivation
Space
Goals
Examples, Mature and Research
Our Research Agenda
The Space of Research
Outline
3
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
1945 1st IBM 
Research Lab 
in NY 
(Columbia U)
Established:        1995 Established:       1972
Established:       1982
Established:       1961
Established:       1998
Zürich
Beijing
Austin
Delhi
Tokyo
Established:      1955
Established:       1995
1952
San Jose
California
Established:      1986
Almaden
Watson
Haifa
IBM Research Worldwide
4
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Geometric growth now generating really large 
quantum gains
Installed base has reached critical mass
Building blocks, painstakingly developed, 
over many years work
Society increasingly accepts & needs I/T
So many more things are now feasible
But, challenges in harnassing I/T technology 
grow; e.g., using massive parallelism
Unabashed Technical Optimism
5
Autonomic 
Computing
6
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
An application 
server typically 
supports
5 Applications
10 EJBs
Hundreds of 
servlets
~ 100 
configuration 
parameters
A web server  
typically serves 
Thousands  
of web 
artifacts
~ 20 
configuration 
parameters
Failure protocols for 
each component are 
different:  time-out, 
number of retries, 
where and what they 
log, how they fail
The increasing challenge of managing large systems is due to the inherent 
complexity of the solution and the sheer number of heterogeneous components
APPC
LU 6.2
SUN
E-mailE-mail
Address 
Capture
AIX
DSS
DSS
Gateways
SUN
Sybase 
Security 
AIX
Sybase
Security 
Servers 
Local
Director
Network
SUN
Sybase Sybase 
Expressnet 
DB Servers
APPC
LU 6.2
APPC
LU 6.2
TPF
TPF
EPRD
SYSPLEXIMSDSUs
PPRD
Complex
IMS
DSUs
IPCE 
SYSPLEX
IMS
DSUs
CICS 
MSC
OS390
OS390
OS390
OS390
CAS
TPF
SYSPLEXIMSDSUs
IPCW 
OS390
Back-end 
Systems
Typical  Enterprise System 
Configuration
Complex System Topology
Messaging 
has ~ 50 
configuration 
parameters
 Front end for online customer service
SUN
App 
Logging
MQ AIX
Logging
MQ
Gateway
Logging
Q
Hub Server Group
Websphere
App Server
Netscape 
Ent. Server
SUN
MQ
HTTP
Presentation    Business Logic                       Gateway
IMSW
IMSS
CAS
MQ
SNA
OICS Engine
AIX
SNA
SNA
DSS
Client
JDBC
HTTP
MQ
SUN
Netscape 
Ent. Server
CIO’s speak out:
“Most of my costs are really pure 
maintenanc  and operations – keeping the 
processes running that keep the ship afloat. 
Our development budget suffers.”  
“Y2K and 9/11 have forced us to look at 
what we have – and we have too much 
complexity.”
7
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Increasing emphasis on Total Cost Of Ownership
Increasing emphasis on QoS
Increasing emphasis on time to market installing 
applications 
Which creates change and instability
Improvement in Manageability
Absolute requirement w/exponential growth of boxes 
outstripping productivity improvements for administrators
Problems:
Increasing complexity
Management is people intensive
Cost of management
Availability of people and skills to do management
Solutions must be open
Industry Trends
8
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Towards Autonomic Computing
Self-optimizing System 
designed to automatically 
manage resources to allow 
the servers to meet the 
enterprise needs in the 
most efficient fashion
Self-configuring 
systems designed to 
define itself "on the fly" 
Self-protecting System 
designed to protect itself 
from any unauthorized 
access anywhere
Self-healing
Autonomic problem 
determination and 
resolution 
9
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
IBM Goals
Create and deploy self-managing 
infrastructure technologies to reduce 
complexity, lower cost of ownership, and 
increase reliability
Establish an architectural framework for 
leadership in Autonomic Computing
Provide technologies to reduce the cost 
of managing systems; that is automating 
automation (automation squared)
10
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
FailureRandom
Malicious
CatastrophicSparse
Aggressive
Load Variability
Attack
Small
Highly malicious
Autonomic Computing Dimensions
Other dimensions
11
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Principles
Local management structure
Redundancy, heterogeneity
Dynamic run-time binding
Validation and self-protection
Requirements
System is always on, always live
Zero IT administration
Any system element can fail
Problems
Testing / verification
Root cause analysis
Global system management
"Evolving" software vs. upgrading
Machine-optimizable components
Standards
Principles, Requirements, Problems
12
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Society
Enterprise
Campus
System
Component
Static, predesigned,
fewer options
Dynamic, self-assembling,
many options
Architectural Styles at Various Stages
13
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
zSeries CPU recovery
CPU duplex
zSeries Sysplex
WebSphere
DB2 self management
Intrusion detection and rejection
Antivirus immune system
Network Dispatcher
IBM Example Mature Technologies
14
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Duplicated:
Complex controls
Arithmetic dataflow
Shared:
Cache controls
Cache data/address flow
R-Unit Check all state updates
Preserve known good state
If error
1. Stop state updates
2. Refresh from saved state
3. Restart CPU
If error persists
1. Extract saved state (SE)
2. Load into spare CPU
3. Start spare CPU
CFW 3/30/00
E-Unit
(unchecked)
Cache
(parity)
I-Unit
(mirror)
E-Unit
(mirror)
R-Unit
(ECC on 
saved state)
I-Unit
(unchecked)
Address
Cache data
Instructions
Results / state updates
Saved state data
zSeries CPU Error Detection and Recovery 
15
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
SMP CEC
CICS
IMS 
DB2 
SMP CEC
CICS
IMS 
DB2 
SMP CEC
CICS
IMS 
DB2 
SMP CEC
CICS
IMS 
DB2 
Sysplex
Timer
Sysplex
Timer
Coupling
Facility
Coupling
Facility
ESCON
Director
ESCON
Director
CICS 
Applications
IMS 
Applications
DB2 
Applications
No SPOF - hardware or software
CEC
16 CPU SMP
Sysplex
32 CECs
or 512 processors
zSeries Parallel Sysplex
16
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Nanny process to restart application server 
processes that have failed or hung.
Basic resource management - threads, 
connections, bean pools allocated as 
needed (within pre-set min and max).
Optimized workload management using both 
session and transactional affinity.
Transaction log recoverability.  
Centralized administration for clustering.  
Can duplicate server configuration across 
servers.  
WebSphere Application Server: Today
17
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Initial Design and Layout
Hardware configuration (a la Estimator for DB2 for 390)
Logical database design
Physical data layout (partitioning, allocation to nodegroups, clustering)
Auxiliary data structures (indexes, ASTs)
Configuration parameters
DB2 for Unix, Windows, & OS/2 V7.1: 
73 database manager parms, 72 database parameters (vs. 52 in V5!)
330 registry variables!
Memory allocation among various heaps, buffer pools, etc.
DB2 for OS/390 and z/OS V7: 
200 DB2 system parameters (ZPARMs) -- 116 hidden
Memory allocation among EDM, Statement Caching, and Sort pools
60 bufferpools with choices of Virtual, Hiper, and DataSpace-backed
 Dynamic Monitoring & Adjustment
 Database statistics to collect and when, Clustering and REORG
 Buffer pool hit ratios, Memory allocation
 Problem determination (deadlocks, bad plans, ...)
 System / query status & visualization of all the above
Huge Scope of DBA Responsibilities
18
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Event Correlation to improve accuracy and 
scalability
Intrusion Tolerance to ensure that the IDS 
itself is protected against attack
Behavior-Based Intrusion Detection to 
enable detection of previously unknown 
attacks
Distributed Event Triage and Correlation
Agent-based ID systems
State of the Art in Intrusion Detection
19
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Automated Virus
Analysis Center
Active
Network
Administratori i
Clientsli
Widget Co.
Analyze
Derive Cure
Distribute
* Sold as Norton Anti-Virus Corporate Edition
Digital Immune System
20
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Automated Virus
Analysis Center
Active
Network
Administratori i
Clientsli
Widget Co.
Wodget Co.
Digital Immune System
21
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Internet
ActiveStandby
Multiple Virtual Clusters
Multiple services within each Cluster
Separate balancing parameters used for each Cluster
Automatically balances load within each Cluster
Fault tolerant: standby ND automatically takes over for failed active ND
Requires no operating system modifications  
Requires no physical alteration to network
Requires no specific code on servers.  
Server agent code can be installed for but is not required
Utilizes up to three metrics to balance within each Cluster
Static: based on counts at ND (no server code)
Advisors: Measures performance of specific application (server code)
System: Measures over all performance of the system (utilizes OS 
performance monitors)
Dynamic feedback used to balance the load
Monitors systems and uses a weighted combination of the metrics to 
reassign load
Weighted round-robin, weights automatically adjusted based on feedback
Remotely manageable
Interfaces available to connect to a broader autonomic system
Start, Stop, Quiesce, machines in a Cluster
Add or Remove Clusters
Layer 3 and layer 7 routing supported
Network Dispatcher: Autonomic Load 
Distribution
22
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
CACHE
eNetDispatcher
CACHE
CACHE
CACHE
CACHE
n
e
t
n
e
t
CACHE
eNetDispatcher
CACHE
CACHE
CACHE
CACHE
Origin 
Server
Origin 
Server
Origin 
Server
     PODs
Front End Cachingi
Origin  i i
caches
Origin  i i
Servers
Content
Management 
Servers
CACHE
HIT
CACHE
CACHE
MISS
ContMgmtSvr
ContMgmtSvr
pre-feed
Content
Sources
Results
Lotus
News/Photos
Publishing
CIS/NetCam
Results
Lotus
News/Photos
Publishing
CIS/NetCam
Four-tier Web Serving Architecture
IBM Olympic Experiences
23
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Oceano 
provisioning and running stateless servers
eWLM 
ebusiness Work Load Manager-open servers  
eBPM 
WebSphere
ABLE 
AI, Policy engine, and Agents
Blue Gene
Cellular computing architecture
Security
Self healing 
Ongoing IBM Research Projects
24
Ongoing IBM 
Research Projects
25
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Requests
Macy's SportsWeb
Macy's
Virtualized Hardware
Single Point of System Management
SportsWeb
Track performance metrics
Aggregate & correlate metrics 
(end-to-end) to SLA violations
Orchestrate reconfiguration 
Fixed resource allocation
Separate management
Best effort basis, using own 
resources
Router
Throttle incoming 
requests 
Océano:
Today:
Océano Project
26
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Self-tuning, End-to-End Performance Management:
Dynamic, allocation of server resources
Workload balancing & routing
Cross platform reporting
Policy based for various classes of users 
& applications
Internet
Appliance 
Servers
Web 
Application 
Servers
Data and 
Transaction 
Servers
Internet/
Extranet
Business 
Partners
Existing 
Business 
Data
  
Distributed Workload Management
27
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Adjust every configuration parameter dynamically,
 while the system is in use!
Expand and shrink memory usage, based on workload
Automatically profile workloads and create/recommend 
indexes, partitioning, clustering, summary tables, ... to 
improve performance
Automatically detect the need, estimate the duration of, 
and schedule maintenance operations 
(like reorg, statistics collection, backup, load, rebind)
Observe actual performance and exploit that 
information to improve operations. 
Recommend action when things aren't they way you 
want them to be.
Project into the future to detect coming problems, like 
low memory or constrained disk space, and notify you 
by page or e-mail days or weeks in advance! 
Wouldn't it be great 
if your database was 
as easy to maintain 
and as self- 
controlled as your 
fridge?
Can your database do this?  Soon it will...
SMART's Vision
28
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Java-based agent framework and AI component library
Agent builder, test and debug tools, multi-agent platform
Add adaptivity through on-line machine learning (data mining)
Policy-based behavior using rules-based knowledge representation
Add reflexive, reactive, and deliberative goal-seeking behaviors
Distributed hierarchical communication and feedback control 
AbleAgent 
Sensors Effectors
Learning
Intelligent 
Control
Reasoning
System
Monitors
System 
Controls
ABLE Autonomic Components
29
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
2.8/5.6 GF/s
4 MB
Chip
(2 processors)
Board
(8 chips, 2x2x2)
Rack
(128 boards, 8x8x16)
22.4/44.8 GF/s
2.08 GB
2.9/5.7 TF/s
266 GB
System
(64 cabinets, 32x32x64)
180/360 TF/s
16 TB
440 core
440 core
EDRAM
I/O
Autonomic Computing Issues: checkpointing, 
routing around failed nodes, data migration, 
communication route optimization
Blue Gene/L System
30
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Behavior-Based Intrusion Detection
Secure Distributed Storage
Secure Boot & System Configuration 
Monitoring
Tamper-responsive hardware
Traps for catching worms and DoS agents
Certified systems that guarantee program 
separation
Current Security Research
31
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Self-managing storage systems
Self-managing data base systems
LEO, DB2 Learning Optimizer
Architecture for control of autonomic 
systems
A Few New Projects
32
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Space Sequential Skip 
Sequential
Random
1
2
3
Device Sequentia
l
Skip Sequ
ential
Random
a
b
c
DatabaseDatabase Autonomic 
Manager
Policy and
History
Policy
Alerts
Storage SystemStorage System autonomic Manager
Policy and
History
File System
File System 
Autonomic 
Manager
Policy and
History
Standard
Porting Layer 
Enhancement
additinos
ALOMS-Tango: Storage for Data Base 
Systems
33
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Statistics
Plan 
Execution
Optimizer
Best 
Plan
l  
ti
ti i rze
t es
laP n Adjustments
SQL Compilation
Actual 
Cardinalities
Estimated 
Cardinalities
1. Monitor
2. Analyze
3. Feedback
4. Exploit
Adjust ts
sti tE a
ar i litiC d na es
t lc
ar i litiC d na es
Learning in Query Optimization
34
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
DataBase
Application and Integration 
Middleware
Operating System
File System
Storage System Processor System
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Managed
Component
Managed
Component
Managed
Component
Managed
Component
Autonomic Manager
Policy based management,measure, model, 
direct
Policy and
History
Policy
Alerts
Measurement
Measurement
Workload and 
service agreements
Workload and 
service agreements
Hints and
Directions
Administrator
Alerts and
measurement
IBM
Managed
Operations
Autonomic Computing - The Whole System
35
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Management
channel
(output)
Management
channel
(input)
Functional
channel
(output)
Functional
channel
(input)
Monitor,
control
Mgt.
Unit
Func.
Unit
Access
control
Encapsulates 
services
Functional unit
Provides the service
Web server, DB, etc.
Management unit
Controls functional unit
Control access
Negotiates for input,
output services
Autonomic System Architecture
An Autonomic Element
36
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Negotiates with
directory for service
Gets location of DB,
storage services
Web ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Web 
Server
Self-configuring
New web server 
added
(Leg of a) Strawman Architecture
An Autonomic System
37
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Web ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Web Server
Self-configuring
New web server 
added
Negotiates with
directory for service
Gets location of DB,
storage services
Negotiates with 
DB,
storage services
(Leg of a) Strawman Architecture
An Autonomic System
38
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Web ServerWeb ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Self-healing
Storage
Storage service dies
(Leg of a) Strawman Architecture
An Autonomic System
39
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
DB gets location of
new storage service
Web ServerWeb ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Self-healing
Storage service dies
Storage
(Leg of a) Strawman Architecture
An Autonomic System (x)
40
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
DB binds new storage
service
Web ServerWeb ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Self-healing
Storage service dies
DB gets location of
new storage service
Storage
DB initializes new
storage service
(Leg of a) Strawman Architecture
An Autonomic System
41
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Web ServerWeb ServerWeb Server
DB
Storage Storage
Systems
Webs of elements
Composition of elements
Composition of services
Late binding
Dynamic
By negotiated SLA
Directory
Self-healing
Storage service dies
DB gets location of
new storage service
DB binds new storage
service
DB initializes new
storage service
Back in business with
no interruption !
Storage
(Leg of a) Strawman Architecture
An Autonomic System
42
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
A long list of difficult problems
Systems
An extremely different way of creating systems
Theory
Difficult issues in complex systems, etc.
Candidate Grand Challenge in Computing 
Research Association (CRA) Grand 
Challenges Conference (ongoing today)
Autonomic Computing:
A Grand Challenge?
43
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Architecture and basic principles
Fundamentals and theory
Standards
Product applications + implications
Software engineering discipline
proof points for all above
(IBM) Autonomic Computing Action 
Framework
44
AZS Presentation to ICS'02 June 25 02                      Copyright IBM
Component System Federation
Optimization 
Algorithms
Data Mining, Continual Optimization
Workload management
Extended Cross system 
workload management
Control Theory Resource SLA managementComponent policy 
management and 
enforcement
Monitoring
Agregating data and keeping 
relevant history
End to End Service level 
agreement 
managementgreement
Distributed Alg. & 
Control
Scripting sensors & control Distributed Alg. & Control
Optimization without 
complete or up to date 
information
Security Intrusion detection Sensor, Instrumentation Federated Intrusion Detection
Special 
Languages
Translate Business Policy to 
component policies
SLA specification language 
and processor,
Policy specification language 
and processor
Rationalizing distributed 
policy
Adaptive/Learning 
Theories
Call Center Optimization,
SLA and Policy Enginex
Complex Systems Automated Operation,Agent Technology,
Autonomic Computing 
framework
Federated 
SystemArchitecture
Infrastructure Component level problem determination,
Unit of work tracking
Time
The Space of Research
45
Thank you for listening.
46