Fall 1999. Distributed/Multiprocessor Operating Systems.

Class synopsis
- just a brief outline of topics covered in class

Class 1: August 23 1999

• Information about: Course Web page, Office Hours, Cheating, Exams, Projects, Lectures, Books, Course outline.
• Mutliprocessor systems:
• Types of machines: UMA, NUMA, NORMA, SIMD, MIMD (possibly SISD, MISD)
• SMP and UMA machines, architecture, operations, speedup
  

Class 2: August 25 199

• Cache Coherence in UMA machine
• Applications for UMA machines
• Operating systems for UMA
• Interrupt handling
• Locking - spin locks

Class 3: August 30 1999

• Usage of Spin locks, mutex locks, interrupt disabling in uni- and multi-processors
• why spin lock not used on single processor
• Interrupt Handling
• Reentrant kernels
• Operating System Structure - Master Slave/Floating Master/Symmetric
• Scheduling
  - preemtion inside spin lock
  - cache corruption
  - context switching
• coscheduling
• Affinity Scheduling
• Handoff Scheduling
• Switched Multiprocessors

Class 4: September 1st

• NUMA architectures
  - scalability
  - architecture details
  - application programming
  - operating systems
  - switch design
• NORMA Architecture
  - relationship to distributed systems
  - switch characteristics
• NORMA programming strategies
  - mesage massing - PVM, MPI, Open MP
  - distributed shared memory
  - tuple space (Linda)

Class X: September 6th - No class - HOLIDAY

Class 5: September 8th

• Distributed Opearting Systems, what are they.
  - The processor pool model
  - The workstation model
• History of distributed systems
  - timesharing systems
  - minicomputer systems
  - networked systems
• Towards centralization of distributed systems
• The Network File Systems
• Dataless workstation configuration
• How NFS works (incomplete)

Class 6: September 13th

• NSF discussion continued
  - Client stubs and server stubs in NSF
  - Remote mounts and file name resolution
  - open/read/write operations
• Client Caching in NSF, file semantics and concurrent access
• Stateless and Stateful file service
• Scalability issues
• Andre File System
  - File server, local files
  - callbacks and coherence control
  - Issues of scalability and cosnsitency in AFS

Class 7: September 15th

• What are are Dist Systems used for?
  -. Distributed Applications
  -. Information Sharing
  -. Resource Sharing
  -. Better Price Performance
  -. Higher Reliability
  -. faster throughput
  -. growth/flexibility/
• What is a Distributed Operating System?
• Integration of autonomous computers.
• Issues:
  - One world view of dist systems
  - Logical and physical distribution
  >> Transparency of:
  *- access,
  *- location,
  *- replication,
  *- failure
  >> Reliability
  *- Fault avoidance,
  *- fault tolerance,
  *- failure detection and
  *- recovery
  >> Flexibility: microkernels
  >> Performance:
  *- caching,
  *- less copying,
  *- low latency protocols,
  >> Scalability
  *- avoid centralized
  *- do things on clients
• - Distributable system models
  - message model
  - object model
  - shared memory model
  

Class 8: September 20th

• Message Passing - what it is
• SEND - the variations in semantics
• RECEIVE - the variations in semantics
• blocking, non-blocking, asynchronous, synchronous
• relative merits of the different kinds
• Programming with send/recv
• global mailboxes - ports
• Client server programs

Class 9: September 22rd

• Client Server Architecture
  - using threads for parallelism
• Marshalling of Arguments
• Dynamic Port Creation
• Port Binding to Name server
• Name server strategies
  - well known ports
  - replications
  

Class 10: September 27th

• Mutltithreaded servers, why we need it:
  - efficiency - even in uniprocessor systems
  - handling recursion
• How to get multithreading?
  - co-routines
  - design of user level non-preemptive threads
  - using semaphores for scheduling and mutual exclusion
• Reentrant Programming and techniques to attain reentrancy
• Kernel level threads

Class 11: October 4th

• More discussion of threads
• Managing servers without threads - using manual programming strategies
• Use of cookies to record state information
• Corrected version of the User-Level threads and server program (uses cookies)
• Taxonomy of threads
• Usage guidelines for threads
• Introduction to Microkernels

Class 12: October 6th

• Mickrokernels - description, usage, facilities and advantages
• Configuring on OS on top of a microkernel
  - services
  - system calls
• Performamnce problems with micorkernels
  - methods of alleviating performance problems
• Message system implementations
  - single machine message passing

Class 13: October 11th

• Message passing between machines
• Local ports and global ports
• Sending to and receiving from ports that are on remote machines
• Network servers, network protocols and inter-kernel messaging
• Access Control, using Capabilities

Class 14: October 13th

• EXAM DAY - no class.

Class 15: October 18th

• Lock management
• Lock types, lock compatibility
• Centralized lock manager for distributed locking (client/server)
• Locking and unlocking read and write locks, with starvation prevention

Class 16: October 20th

• Upgrading of read locks to write locks
• Semantics of upgrading
• Deadlocks due to upgrading
• Deadlock resolution
• Starvation due to deadlock resolution
• Starvations due to upgrading (not discussed - open problem)
• Project details

Class 17: October 25th

• MILAN (Calypso, Chime) , by Tom Boyd

Class 18: October 27th

• Computing Communities by Tom Boyd

Class 19: November 1st

• Shortcoming of Message passing as a programming method
• Remote Procedure calls (RPC)
• The IDL compiler
• Client stub generation and how client stubs work (marshalling, type checking, parameter types)

Class 20: November 3rd

• RPC Continued...
• Server stub generation
• How the RPC server and clients communicate
• Name services
• Type Checking
• SunRPC, DCE RPC, MS RPC
• What is DCE

Class 21: November 8th

• Introduction to DSM
• Shared memory semantics (coherence, threads, processes, sharing)
• Distributed Shared Memory - approaches
• Page fault handling and DSM servers
• Invalidations
• Algorithms for implementing a sequential consistent, page-based DSM system

Class 22: November 10th

• Performance problems of DSM
    - Page Shuttling
    - False Sharing
• Semantics of DSM systems
• Strict consistency, Sequential Consistency, Causal Consistency, PRAM, Weak and Release Consistency
• What is release consistency (acquire/release of shared memory)
• How release consistency reduces page shuttling and false sharing

Class 23: November 15th

• Release consistency implementation
  - page diffs
  - multiple locks
  - why is performance good with RC
• performance of DSM systems
• The Muddy Children Problem
• Hierarchies of knowledge
• The definition of "Common Knowledge"

Class 24: November 17th

• PROJECT 2 handed out. Please note STRICT due date.
• Consensus and agreement in distributed systems
• The Coordinated Attack Problem
• The 2-phase commit algorithm
• Introduction to "time"

Class 25: November 22nd

• Discussion of Project 2 requirements
• Concept of time and clocks
• Real time clocks
• Clocks synchronization for real time clocks
• Modeling clocks as a counter
• Modeling time as event ordering
• Lamport's clock criteria

Class 26: November 24th

• Project 2 web pages have been updated
• Lamport Clocks
• Distributed Mutual Exclusion

Class 27: November 29th

• Lamport Solution to DME
• Other solutions, Maekawa style solutions
• Centralized solutions - comparisons
• Distributed Snapshots

Class 28: December 1st

• Distributed Snapshots
• The Algorithms
• The Sample programs and examples of state recording
• The properties of recorded state
• Informal proof of correctness

Class 29: December 6th

• Project discussion, Snapshots discussion
• Distributed deadlock detection (edge chasing algorithm)
• Fault tolerance in distributed systems
• Data Fault Tolerance [Replication Management]
• Read 1 Write All, Available Copies, Quorum Consensus, etc.

Class 30: December 8th

• LAST CLASS

Final Exam: Tuesday December 14th 10am - 11:50am