The document discusses goals for a dissertation on monitoring, diagnosing, and repairing systems. It aims to reduce the cost of system administration by automatically detecting and repairing problems faster. Key innovations proposed include using replicated, semi-hierarchical databases for fault tolerance and scalability, self-describing data structures, end-to-end notification, aggregation and high-resolution displays to reduce information, self-configuring systems, and secure remote actions. The system will be evaluated through fault injection, feature comparison, real-world usage, and testimonials.

1. Monitoring, Diagnosing, andMonitoring, Diagnosing, and
RepairingRepairing
Eric Anderson
U.C. Berkeley

2. 2Jun 28, 2013
OverviewOverview
 What is System Administration?
– What is the problem?
– Goals of Dissertation Research
– Goals of System Administration
 Monitoring, diagnosing, and repairing
 Dissertation Timeline
 Conclusion

3. 3Jun 28, 2013
What is the problem?What is the problem?
 Problems occur in systems, and result in loss of
productivity
– Server failures  denial of service
– System overload  lower productivity
 Cost is too high
– Cost of ownership estimated at $5,000-$15,000/year/machine
– Median salary (~50k) / (median # machines/admin)  $700
 Our goal: Reduce cost by
– Repairing problems faster (possibly automatically)
– Handling more problems

4. 4Jun 28, 2013
Goals of Dissertation ResearchGoals of Dissertation Research
 Describe field of System Administration
 Monitoring, Diagnosing, and Repairing:
– Approach: Synthesize solutions from other fields of research
1) Detect previously ignored problems
2) Automatic repair of some problems
3) Reduce number of administrators needed
4) Support users’ understanding of system
 Apply here & distribute software
 Thesis: Through our approach, we can achieve
goals 1-4.

5. 5Jun 28, 2013
Goals of System AdministrationGoals of System Administration
Goal: Support cost-effective use of the computer
environment
More specifically (some non-technical):
Environment: uniform, customizable, high performance and
available
Faults & errors: recovery from benign errors, protection from
malicious attacks
Users: training, accounting & planning, legal

6. 6Jun 28, 2013
Monitoring, Diagnosing, andMonitoring, Diagnosing, and
Repairing (MDR)Repairing (MDR)
• Introductory examples
• Fundamental requirements
• Environmental constraints
• Previous work
• Six key innovations
• Architecture
• Details on innovations
• Evaluation methodology

7. 7Jun 28, 2013
MDR: Examples — IntroMDR: Examples — Intro
 Four examples
1) Broken component
2) Resource overload — transient
3) Resource contention — user program
4) Resource exhaustion — long term
 Previous Solutions
– Pay someone to watch
– Ignore or wait for someone to complain
– Specialized scripts (not general  vast repeated work)

8. 8Jun 28, 2013
MDR: Example 1MDR: Example 1
Web server has crashed/hung
 Gather information: process existence, service
uptime, restart times
 Analyze data: process not responding, and hasn’t
been recently restarted.
 Automatic repair: restart daemon.
 Notify administrator: had to restart daemon.

9. 9Jun 28, 2013
MDR: Example 2MDR: Example 2
The NOW is “slow.”
 Gather data: load, process info, CPU info
 Analyze data: bounds on expected values
 Notified administrator: fileserver overloaded.
 Visualize data: nfsd’s are overloaded.
 Repair: admin moves data, adds disks, or starts
more nfsd’s

10. 10Jun 28, 2013
MDR: Example 3MDR: Example 3
User running program
 Gather: user statistics, CPU, disk
 Visualize: spending too much time waiting on remote
accesses
(User fixes program, gathering, visualization repeated)
 Analyze: some nodes have less throughput
 Visualize: those have other jobs running on them
 Repair: user is benchmarking so kills all extraneous
processes

11. 11Jun 28, 2013
MDR: Example 4MDR: Example 4
Web server increasing beyond capacity
 Gather: CPU, request rate, reply latency
 Analyze: Burst lengths getting longer, latency
increasing
 Visualize: Graph of burst lengths & CPU usage over
time
 Repair: Order more machines, install load balancer

12. 12Jun 28, 2013
MDR: Fundamental RequirementsMDR: Fundamental Requirements
• Gathering
• Flexible data gathering, self-describing storage
• Analyzing
• Calculate statistical measures, identify relevant statistics.
• Notifying
• Flexible infrequent messages to administrators or users
• Visualizing
• Maximize information/pixel, support multiple interfaces
• Repairing
• Automate simple repairs, support group operations

13. 13Jun 28, 2013
MDR: EnvironmentalMDR: Environmental
ConstraintsConstraints
 Change is inherent
– Lack of Web/Mbone 5 years ago, now most/many have these.
 Problems on many time-scales
– Second-Minute transients vs. Week-Month capacity problems
 Must operate under very adverse conditions
– Often used when system is broken
– Would like at least post-mortum analysis
 Need to handle hundreds – thousands of nodes
– Scalability: All sites are getting larger, possibly wide area
– Our system has 200 (NOW) – 2000 (Soda) nodes

14. 14Jun 28, 2013
MDR: Previous SystemsMDR: Previous Systems
 Many previous systems: I’ve looked at about 16.
 Not comprehensive, not extensible.
 Look at a few that did a nice job of a piece:
 [Fink97] — Run test, notify display engine
+ Easy to add tests
+ Selectivity of notification good
– Tests are just programs (redo gathering)
– Central, non-fault tolerant solution
– Many hard coded constants

15. 15Jun 28, 2013
MDR: Previous Systems, cont.MDR: Previous Systems, cont.
 [Hard92] — buzzerd: Pager notification system
+ Flexible rules for notification
+ External interface for adding notify requests
– Simplistic gathering
– Poor fault tolerance
 [Pier96] — Igor group fixes
+ Flexible operations
+ Nice reporting of success/failure
– Weak security, runs as root
– No delegation of responsibility

16. 16Jun 28, 2013
MDR: Six Key Innovations (1-3)MDR: Six Key Innovations (1-3)
 Replicated, semi-hierarchical, data storage nodes
– Rendezvous point for programs
– Handles scaling and fault-tolerance
 Self describing structures
– Functions (visualize, summarize) + data go in database
(OO)
– DB has machine and human readable descriptions of data
 End to end notification
– Detect problems in MDR system
– Guarantee important messages get to users

17. 17Jun 28, 2013
MDR: Six Key Innovations (4-6)MDR: Six Key Innovations (4-6)
 Aggregation and High Resolution Color Displays
– Reduce information to manageable amounts
– Maximize information per unit area
 Partially self-configuring
– Learn averages, deviations, burst sizes
– Learn which values are relevant to problems
 Secure, user-specified group repairs
– Don’t enable malicious attacks
– Automate repairs of many machines

18. 18Jun 28, 2013
MDR: ArchitectureMDR: Architecture
SQL-based Data Repository
Gather Agent
vmstat thread
ping thread
tcpdump thread
Diagnostic
Console
E-mail or
Phone
Notifier
Long-term
graphing Tolerance,
Relevance
Learner
Daemon
Restarter
Aggregation
Engine

19. 19Jun 28, 2013
MDR-Arch: DerivationsMDR-Arch: Derivations
SQL-based Data Repository
Diagnostic
Console
E-mail or
Phone
Notifier
Tolerance,
Relevance
Learner
Daemon
Restarter

20. 20Jun 28, 2013
Key: Semi-Hier. DBs.Key: Semi-Hier. DBs.
 Fault tolerance
 Scalability:
– Caches don’t need to commit to disk — authoritative copy
elsewhere.
– Batching updates over wide area links.
Top level cache Top level cache
Mid level cache Mid level cache Mid level cache
Per-node
database
Per-node
database
Per-node
database
Per-node
database
Per-node
database

21. 21Jun 28, 2013
Key: Self-DescribingKey: Self-Describing
 De-couple data gathering, data storage, and data use
 Self-Describing for Humans
– Descriptions of meanings of values stored with tables
– Description of methods of gathering stored with tables
– Column names help with self
 Self-Describing for Computers
– Functions for visualizing or summarizing data
– Indication of resource selection from resource statistics

22. 22Jun 28, 2013
Key: End-to-End NotificationKey: End-to-End Notification
Recall: System must operate under extreme conditions
 Humans must validate that system is still working
– Standalone display can indicate timestamps, mark out of
date data
– Wireless machine could intermittently contact notification
system
– Pager could be automatically paged every so often
 Problems should be propagated to end users.
– Flexible notification — connected systems, e-mail, pager.
– Limit over-notification

23. 23Jun 28, 2013
Key: Aggregation & HiResKey: Aggregation & HiRes
 System target has hundreds – thousands of nodes
 Aggregate by showing out of bounds, relevant values
(via automatic tuning)
 Also want overview of system
– Aggregate across similar statistics; show value (fill) &
dispersion (shade)
– Use color to highlight important values.
– Aggregate across values (machine utilization = CPU + disk +
memory)
– Maximize data/pixel [Tufte]

24. 24Jun 28, 2013
Key: Agg & HiRes: SnapshotKey: Agg & HiRes: Snapshot

25. 25Jun 28, 2013
Key: Self-ConfiguringKey: Self-Configuring
 Single statistics
– Phase 1: Calculate averages, standard deviations, burst
sizes
– Worked in other systems [Jaco88, Karn91]
 Identify relevant statistics
– Give system Boolean examples (variables out of bounds,
and system working/not working) get function.
– Works for Boolean disjunctions in some cases:
• With lots of irrelevant variables [Litt89]
• With random bad examples [Sloa89]
• In some cases, with malicious bad examples [Ande94]

26. 26Jun 28, 2013
Key: Secure Remote ActionsKey: Secure Remote Actions
 Security because of malicious attacks, benign errors
 Delegation to remove SA from the loop
 Independence from particular algorithms
– Building a library
– Program with principals (hosts, users), and properties
(signed, sealed, verifiable)
 Use secure, run-time extensible languages
 Actions report through gathering system

27. 27Jun 28, 2013
MDR: Testing MethodologyMDR: Testing Methodology
 Fault injection
– Deliberately make the system slow
– Break hardware/software components
 Feature comparison
– Paper comparison with other systems
 Usage in practice
– Experience important to show system works
– We have need of administrative tools
 Testimonials
– Experience at other sites lends credibility

28. 28Jun 28, 2013
MDR: DemoMDR: Demo
 Hierarchical structure working (1 level right now)
 Alternative Interface
 Fault Injection
 Need for Aggregation
 Crufty right now
 Demo

29. 29Jun 28, 2013
Timeline: Key PiecesTimeline: Key Pieces
1) (DBs) Replicated, semi-hierarchical, data storage nodes
2) (SDS) Self describing structures
3) (Vis) Aggregation and High Resolution Color Displays
4) (E2EN) End to end notification
5) (ReS) Automatic Restart
6) (Cfg) Partially self-configuring
7) (Rep) Secure, user-specified group repairs

30. 30Jun 28, 2013
TimelineTimeline
Deadlines:
June, 1997 Dec, 1997 Dec, 1998June, 1998
LISA 6/97 USENIX 12/97 OSDI 3/98 Graduation 12/98
Prototype 1,2,3
(DBs, SelfD, Vis)
Prototype 4,5
Notify, Restart
Prototype 6,7
AConfig, Repair
LISA 6/98
Experience
with 1-7
SOSP
3/99
Architecture of
Complete System
Writing
Mar, 1999

31. 31Jun 28, 2013
ConclusionConclusion
 Description of field shows breadth
 Monitoring, diagnosing, and repairing shows depth
– Examples show importance of problem
– Fundamental goals & environmental constraints show
understanding of problem
– Key innovations show differences from previous systems.
– Architecture and initial prototype show approach to problem
– Testing methods show ways to validate solution.
 Timeline shows plan & milestones to graduation

32. Old SlidesOld Slides

33. 33Jun 28, 2013
SolutionsSolutions
 Managing stable storage
 Supporting users
 Simplifying security
 Monitoring, diagnosing, and repairing

34. 34Jun 28, 2013
Managing Stable StorageManaging Stable Storage
 Consistency vs. availability
 Fault tolerance
 Scalability
 Recoverability
 Customization

35. 35Jun 28, 2013
Supporting UsersSupporting Users
 Automated help desk
– Searchable collection of questions
– Easy method for addition
 Remote device access
 Site-wide training

36. 36Jun 28, 2013
Goals: EnvironmentGoals: Environment
 Uniform
– Supports user mobility by eliminating arbitrary changes
– Increases effectiveness by avoiding need for users to learn multiple
interfaces
 Customizable
– Handles special systems and special needs [firewalls, servers]
– Obviously reduces uniformity

37. 37Jun 28, 2013
Goals: Environment, cont.Goals: Environment, cont.
 High Performance
– Increases effectiveness of users [HCI/psych]
– Limited by cost-effectiveness
 Available
– Effectiveness is 0 if system isn’t working
– Balanced against expense

38. 38Jun 28, 2013
Goals: Faults & ErrorsGoals: Faults & Errors
 Benign errors:
– Accidentally deleted files
– Unnoticed runaway processes
 Malicious attacks:
– TCP SYN attack
– Sendmail bugs
– Data stealing
– False data injection

39. 39Jun 28, 2013
Goals: UsersGoals: Users
 Training
– Troubleshooting = one-on-one training
– Larger sessions = classes
 Accounting
– Supports management, helps billing
 Capacity Planning
– Expanding systems takes time
 Legal
– Sensitive information needs protection

40. 40Jun 28, 2013
Simplifying SecuritySimplifying Security
USENIX talk says “If cryptography is so great, why isn’t it used more?”
SA’s worry about security to protect data.
 Goal: Ease development of secure applications
 Write programs using principals & properties rather than keys and algorithms
 Unify various forms of available cryptography (public key, secret-key, PGP,
Kerberos)
 My use: protected, transferable rights to allow various actions
– Modify system configurations (add filesystems, printers)
– Kill/restart processes (runaway, after configurations modified)
– Access data (private logs, for backups, etc.)

41. 41Jun 28, 2013
ConclusionConclusion
 System administration as area of research
– Description of field
– Areas for future research
• Managing stable storage
• Supporting users
 Initial investigation of research area
– Monitoring, diagnosing, and repairing
• Broad, draws from many fields