Mach System - Operating System - Lecture Slides, Slides of Computer Science

Download Mach System - Operating System - Lecture Slides and more Slides Computer Science in PDF only on Docsity!

1

The Mach System

2

Overall Mach Structure

• Additional, middle,

layer of operating

system

• Other O/S’s typically

implemented as user

level O/S’s

Mach microkernel

application level O/S (e.g., UNIX)

User application

5

• http://developer.apple.com/documentation/Darwin/Concep

tual/KernelProgramming/Architecture/chapter_3_section_ 1.html

Figure 3- Mac OS X kernel architecture (4.4 BSD)

7

Goals of Mach

• “Simple” kernel
  • Only one way to accomplish kernel programming tasks
• Diverse hardware support
  • “Multiprocessing”: Tightly coupled or loosely coupled
    • Varying degrees of shared memory access
  • One to thousands of processors
  • Heterogeneous system support
• Distributed & object-oriented
  • All kernel objects within the system are location independent
• Integrated memory management & interprocess

communication (IPC)

• Varying network speeds
• Extensible microkernel Docsity.com

9

10

Interprocess Communication

• Ports
  • Microkernel protected communication channel
  • Communication occurs by sending messages to ports
  • Microkernel object reference mechanism
  • Provides access to devices, tasks, threads, memory objects, and processors
  • Allow microkernel objects to transparently reside anywhere in network
  • Threads have port rights (send & receive)
• Port sets
  • Group of ports sharing a common message queue
  • By receiving messages for a port set, a thread can service multiple ports
  • Similar to select call in BSD Unix

12

“User-Level” Message Server:

NetMsgServer

• Networking daemon
• Enables location-independent naming of ports
  • Does not matter which computer a port is on
  • NetMsgServer dynamically resolves the addresses
• Services
  • Forwards message between computers
  • Network wide name service
    • Tasks can register ports for lookup by tasks on any other computer on network
  • Distributed database of port rights
  • Tracks all rights & out-of-line memory passed in intercomputer messages & arranges for transfers
  • Data conversions (different computer architectures)
  • Proxy ports for passing port rights between computers

14

Mach Messages

• Data in messages: typed
• Simple types (integer, float etc)
• In-line and out-of-line data
• Out-of-line data: Pointers
  • Kernel interprets these pointers as part of message passing
  • Receiver on same computer
    • Don’t want to necessarily copy message from sender to receiver - Takes time to copy messages
    • Instead, when message contents unchanged, use virtual memory-based technique for improving efficiency - A kind of shared memory solution - “copy-on-write”
  • Receiver on different computer
    • Need to copy data referred to by pointer Docsity.com

16

Mach Memory Objects

• Memory objects: objects in memory mapped from devices
  • Ports provide operations
  • Transparent remote & shared access
• “may be backed by user-level memory managers, which take the place of more traditional kernel-incorporated virtual-memory pager found in other operating systems” (p. 874-875)
• For user processes (i.e., not O/S user-level processes), usually mapped from a disk device - Can implement access to files in this way - Map a series of disk blocks onto memory
• Can also map other devices to memory
  • Gives “memory mapped I/O”
    • E.g., Writing to a memory address writes to the control register of a device; reading from a memory address reads from the control register of a device

17

User-Level Memory Managers

• May reject the kernel’s replacement victim if it

knows of a better candidate

• May monitor the memory object it is backing, and

request pages to be paged out before memory

usage invokes Mach’s pageout daemon

• Can control order of operations on secondary

storage to enforce consistency constraints

• Control memory mapped file access

19

Mach I/O System - 1

• Device management
  • Memory Objects used for mapping devices to memory
    • handled in ‘user level’ processes
  • Non-memory mapped devices handled in microkernel
• Location-transparent device management
  • Devices are Memory Objects
  • Ports on these devices can be accessed by any task with appropriate rights
• “User level” I/O can increase performance
  • Avoid data copy between kernel memory space and user level memory space
  • In experiments, performance with “user level” device management at least as good as kernel-level - In some cases performance increased

Mach I/O System - 2

• Functionally grouped device drivers
• Screen, console, disk, tape, serial line, Ethernet
  • e.g., Instead of character & block devices
• Each driver
  • layer(s) of chip independent code
  • minimal device dependent code: in microkernel
  • Example: serial line driver
    • Device independent components
      • open, close, read, write
      • start, stop
      • modem controls
      • interrupt handlers for devices that operate on one-interrupt per character basis
      • send/receive char– polling mode
    • Device dependent components
      • Only operations that manipulate device registers directly 20

22

Process Management

• Task
  • Contains threads
  • Virtual address space + Port rights ( capability )
• Threads
  • Running : executing or waiting to be allocated to a processor - Considered to be running even when blocked within kernel (e.g., on a page fault)
  • Suspended : Neither executing on a processor nor waiting to be allocated to a processor. - Used to achieve a type of thread-level synchronization
• CPU scheduling

23

Mach CPU Scheduling

• Global & local (single CPU) run queues
  • Thread assigned to processor from a run queue
  • Only threads are scheduled; compete equally for CPU
• Threads have dynamic priorities 0-
  • Generally equal competition for resources
  • Priority based on exponential average of CPU usage
    • Reduces priority with increasing CPU usage
• Threads can be “bound” to a CPU
  • Thread will only run on that CPU
  • E.g., for a thread that handles a device associated with that CPU
• Variable time quantum (processor time per process)
  • Dependent on # processors & # processes