Week 03 Notes for CS 3750 -- Fall 2001

• Take Roll.
• Look at the upcoming schedule and homework.
• Announcements
  • I'm trying to decide between teaching Operating Systems II and Theory of Algorithms in the spring.
• Chapter Three Lecture Ideas
  • For a personal computer user, what are the advantages of multiprogramming?
  • How does Unix accomplish these aspects of process control?
    • process creation and destruction?
    • suspension/resumption
    • synchronization
    • communication
    • deadlock handling
  • do a "ps" on a Unix system and discuss interesting processes that are running (probably check processes that are vital to the running system first, then other processes owned by root.).
  • Discuss generally how the OS could do the following tasks connected with file management:
    • create/delete files
    • create delete directories
    • support file access primitives
    • map files to secondary storage
    • back-up files in RAM onto stable storage
  • Discuss generally how the OS could do the following tasks connected with disk management:
    • free-space management
    • storage allocation
    • disk scheduling
    (Management of this bottleneck is critical. One of the big developments of Unix was the re-design of the file system. It greatly improved throughput.)
  • Discuss these basic OS services:
    • program execution
    • I/O operations
    • File-System manipulation
    • Communications
    • Error detection
    • Resource allocation -- something like a request/release paradigm is necessary in a multiprogramming environment.
    • Accounting
    • Protection
  • System Calls
    • the operating system services mentioned above are furnished by means of OS system calls.
    • If we think of the OS as a data structure, then system calls are the operations of the data structure.
    • The "system call instruction" generates a trap (interrupt).
    • System call parameters may be left in registers, RAM, or program stack. Registers can hold pointers to areas in RAM containing parameters.
    • Look on page 66 at the list of system call types. Do we understand what all these operations do?
  • Discuss "single-tasking approach" of DOS versus "multi-tasking" approach of Unix.
  • Any questions about the "message passing" and "shared memory" communication paradigms?
  • Discussion of shells?
  • Operating System Structuring
    • Some OS's are not very modular
    • Some are built in layers, but the layers can "bypass" each other.
    • Some are more strictly layered -- bypassing is not common. Each layer depends on the one underneath.
    • layered design facilitates testing.
    • It's not easy to design layers without any "circular" dependencies -- (e.g. backing store driver depends on CPU-scheduler and vice-versa.)
    • Microkernel design puts just a few essential services (say process & memory management plus process communication) in the kernel and the rest is pushed to "user-level"
    • With microkernel architecture it becomes fairly easy (if not maximally efficient) to emulate many different operating systems "on top of" the microkernel. The idea is to interpret foreign system calls with microkernel system calls.
  • The virtual machine idea is to implement in software "an interface that is identical to the underlying bare hardware." -- nice way to isolate users -- secure -- each user can run a different operating system. A user can experiment on a new operating system design without risking bringing down the part of the system being used by others. The Java virtual machine is a modern instance of this type of design. The IBM version dates back to the late sixties.
  • OS Implementation Principles
    • Separate policy from mechanism.
    • Write most of the OS in a high level language. (one writes code faster; code is more compact; code is easier to understand and debug; the operating system is "portable." -- There is really no downside -- modern compilers generate efficient assembly code. Increases in efficiency of OS usually comes by using better data structures and algorithms -- we can write a few small pieces of the OS in assembly where that is really needed for efficiency or for compatability with the hardware.)
  • System generation -- setting compile parameters for the OS versus table driven systems that are pre-compiled. The trend is toward pre-compiling.
• Chapter Four Lecture Ideas
  • What is a process?
  • What is a thread?
  • What is process state?
  • What is the relation of "process state" to the term "state" as used in science?
  • What is a process control block?
  • What is process context?
  • What information should be in a process control block? Relation to what is needed to restart the process?
  • Discuss ready queueue
  • Discuss device queues
  • Discuss representing a computing system as a network of queues -- you can get a lot of good insight from studying characteristics of the network of queues.
  • Discuss how the lifetime of a process is described as a journey through a queueing network.
  • Discuss job pools and long term scheduling in batch systems
  • Discuss short-term scheduling in multiprogramming systems.
  • What is medium-term scheduling?
  • Discuss the importance of having a good job mix to keep the CPU and I/O channels busy. (I/O-bound vs CPU-bound jobs) This is something the scheduler must consider.
  • What are context switching and swapping -- important to understand the difference and to use the terms correctly.
  • Discuss fork, (forms of) exec, wait, exit and the relation to the functionality we see in the unix shell. -- Note the OS serves as an intermediary between parent and child processes, allowing them to synchronize some of their activities.
  • Discuss the "shared-memory solution of the bounded buffer problem" -- the producer-consumer problem. Note that busy waiting is used -- effective if process time in CPU is limited by a timer, or if there is more than one CPU, but wasteful of CPU time. Note that one slot in the buffer is "reserved". It's a "trick" question to ask how to remove this restriction!!
  • Communication between processes can be done with shared memory, but also can be done using an IPC facility furnished by the OS.
  • Message passing of the OS can be implemented with shared memory, hardware bus, network, or ... there has to be some type of communication link.
  • Discuss Mach and/or Windows 2000 message passing
  • Discuss client/server communication via socket interface.
  • Discuss how RPC abstracts the procedure call mechanism so that a process on machine X can in effect make a function call that is carried out by a process on machine Y connected to X by a communication link. -- Typical implementation is for a client on X to connect to an RPC server daemon on Y. The RPC daemon spawns a process that executes the function and returns the results to the daemon, which then send the results back to the client on X. Stubs provide the "glue." Difference in machine representation of data is an issue. Timestamps can be used to prevent duplication of procedure call if RPC message is duplicated on the network.
  • One may implement a distributed file system by using RPC for doing the file operations like read, write, rename, delete, and so forth.