InfoMagic Source Code 1993 July

home *** CD-ROM | disk | FTP | other *** search

/ InfoMagic Source Code 1993 July / THE_SOURCE_CODE_CD_ROM.iso / mach / doc / notes / micro-kernel.ibm.Z / micro-kernel.ibm

Wrap

Internet Message Format | 1993-01-10 | 18.2 KB

Date: Mon, 11 Jan 93 12:38:31 EST This came off some netnews board and is supposed to be from IBM's OS/2 Developer bboard.... After the first few screenfuls it gives a pretty good overview of the Mach 3.0 micro-kernel. -------------------------------------------------------------------- M I C R O K E R N E L T E C H N O L O G Y "A modular, portable software kernel promising powerful open systems advantages for system developers and end users." I B M MICROKERNEL TECHNOLOGY Dramatic advances in computing architecture, hardware designs,and technology have spawned an amazing variety of programming methods, user interfaces, development tools, communications techniques and application software. Delivering previously unheard of advantages to end users in performance, service level, and system utility, the overwhelming diversity of system offerings has also heralded the cry for "open systems" & "standards". IBM, long a leader in creating new systems delivering the broadest range of application capabilities to the widest range of users, is again committed to developing the new, open, technologies needed for the next computing decade. IBM developers have begun to prototype the next generation of highly portable systems software, and related applications, around a microkernel. A microkernel provides a new way of structuring systems software to reduce it's complexity and to increase it's portability. It presents the potential for building a single, truly open, computing environment: using a microkernel "core" product executing multiple diverse operating systems as "applications" on a single hardware platform. The promise of microkernel-based system software lies in providing better ways to cope with the increasing complexity of system designs, the potential for systems software to be more "in tune" with the startling leaps forward in processor hardware, and ultimately, the ability of the end user to select the best available application... independent of the available processor. Systems built using microkernel technology will deliver a new standard for modular, portable, operating systems development platforms. Rather than creating a single, monolithic system, developers create a set of servers that can be configured in a variety of ways to provide the target operating system. (1) One of the most discussed advances in operating systems in recent years, is the microkernel - a small, message-based piece of systems software running in the most privileged state of the computer, supporting the rest of the operating system as a set of server applications. Interest in microkernels has grown as system developers have reacted to the complexity of current operating system implementations and as the research community has demonstrated the feasibility of the microkernel concept. The Mach microkernel technology developed at Carnegie Mellon University serves as the basis for IBM's microkernel work. On the Mach base IBM is experimenting with new ways of implementing low-end environments, developing stand-alone file servers, integrating multiple operating system personalities on a single computer, and enabling new ways of delivering DOS execution environments. By virtue of their size and their ability to support standard programming services and features as application programs, micro- kernels themselves are simpler than standard operating systems. With a microkernel, operating system function is broken down into modular pieces that may be configured in a variety of ways, permitting larger systems to be built by adding pieces to smaller ones. Microkernels also make it easier to support multiprocessors by creating a standard programming environment that can use multiple processors if they are present but does not require them. The specialized code for multiprocessors is restricted to the microkernel itself. Moreover, networks of communicating microkernels can be used to provide operating system support for the emerging class of massively parallel machines. Since they are small and have relatively little kernel- level code to execute, microkernels provide a convenient way to support the real-time features needed for multimedia, device control and high speed communications. Finally, well structured microkernels provide an insulating layer for hardware differences that are not masked by the use of higher level programming languages. Thus, they make porting code easier and increase code reuse. Current Mach research and development activities include on-going research work at Carnegie Mellon, as well as Mach-related work at the University of Utah, the University of Arizona and Cornell University. The Open Software Foundation Research Institute, building on it's previous work and research done at Carnegie Mellon has created a version of Mach that supports massively parallel supercomputers. There is also work underway at the Center for High Performance Computing of Worchester Polytechnic Institute on real-time extensions and file system servers for a Mach environment. (2) +------------------------------+ +------------------------------+ | | | | | | | | | Dominant Personality | | Secondary Personality | | Applications | | Applications | | | | | | (MIC) | | (MIC) | +------------------------------+ +------------------------------+ ================================================================== "True Application"/Server Boundry ================================================================== +------------------------------+ +-------------+ +-------------+ | | | | | | | Dominant Personality | | Secondary | | Secondary | | Server | | Personality | | Personality | | | | Server | | Server | | (MIC) | | (MIC) | | (MIC) | +------------------------------+ +-------------+ +-------------+ ================================================================== Personality/Personality-Independent Boundry ================================================================== +--------+ +--------+ +----------+ +---------+ +----------+ | | | | | | | | +-+--------+ | | Master | | File | | Hardware | | Default | | Device | | | Server | | Server | | Resource | | Pager | | Drivers | | | | | | | Manager | | | | | | | (MIC) | | (MIC) | | (MIC) | | (MIC) | | (DDC) |-+ +--------+ +--------+ +----------+ +---------+ +----------+ ================================================================== User/Privileged Boundary ================================================================== +-----------+----------------------------------------+-----------+ | | Enhanced Mach 3.0 Microkernel | | | | | | | | | | | | | Virtual | IPC | I/O Support | Host and | Tasks & | | Memory | | & Interrupts | Processor Sets | Threads | | | | | | | | (MIC) | (MIC) | (MIC) | (MIC) | (MIC) | +-----------+-------+---------------+----------------+-----------+ | (MDC) | (MCD) | (MDC) | (MDC) | (MDC) | +-----------+-------+---------------+----------------+-----------+ (MIC) - Machine-Independent Code (DDC) - Device-Dependent Code (MDC) - Machine-Dependent Code (3) THE MACH ARCHITECTURE The functions performed by the Mach microkernel itself are limited to reduce its size and to maximize the amount of code that runs as an application program. The microkernel includes only those functions that are required to define a set of abstract processing environments for application programs and to permit applications to work together to deliver services and to act as clients and servers. As a result, the microkernel provides only five different types of services. * virtual memory management * tasks and threads * interprocess communications (IPC) * I/O support and interrupt management * host and processor set services Other operating system functions such as file systems, device support and traditional programming interfaces are placed into one or more server application programs. The enclosed diagram illustrates this function split by showing how IBM is using Mach to support an experimental low-end version of AIX* for PS/2* and other Intel386* and Intel486* based hardware. MACH TECHNOLOGY OVERVIEW The virtual memory component of the Mach microkernel supports large, paged, sparse address spaces that are composed of memory objects. Each memory object is managed by a memory pager, a task outside the kernel that provides the backing storage for the pages of the memory object. Address spaces are manages by mappings or allocating memory objects within them. Mach manages memory protection and sharing on a memory object basis in an abstract way, independent of any particular processor address translation hardware. In particular, Mach makes heavy use of copy-on-write to permit programs to share memory objects without copying large numbers of pages when a new address space accesses the memory object. New copies of the page are created only when a program in one of the address spaces updates them. When Mach takes a page fault in a memory object and does not have the page in memory or when it has to remove pages from memory due to the requirements of other programs executing on the machine, it notifies the pager for the memory object in which the fault occurs using the IPC mechanism. It is then up to the pager, an application server, to determine how to provide or to store the data. This permits the system to define different semantics for the memory objects based on the needs of the programs that use them. The default pager shown in the diagram provides a pager for the memory objects that do not require anything other than the usual semantics of virtual memory. (4) Mach manages the execution environments for the programs by providing multiple tasks and threads. Each Mach task has it's own address space, or map. It assigns memory objects that the task has mapped to ranges of addresses within the address space. The task is also the unit of resource allocation and protection in Mach with tasks being assigned capabilities and access rights to the IPC facilities of the system. In order to support parallel execution by a program within a single address space, Mach separates the execution environment from the actual running of streams of instructions. The streams of computation including the processor resources needed to support them are called threads. Thus, a program can be loaded into a Mach task and can be executed in several different places in the code at the same time on a multiprocessor or parallel machine. This can result in better application performance. The Mach IPC system provides the basic mechanism that allows threads running in different tasks to communicate with each other. The IPC system supports the reliable delivery of messages on ports. Ports are protected channels between tasks. Each task that uses a port is assigned a set of rights to that port. These rights may be different for different tasks. Only one task can receive on any port although any thread within the task may execute the receive operation. One or more tasks may hold rights to send to a port. The kernel permits tasks to use the IPC system to transfer port rights among themselves. It also provides a high performance way of passing large data areas in messages. Rather than copying the data, the message contains a pointer to it: this is called a pointer to out-of-line data. When the kernel transfers the message from the sender to the receiver, it makes the memory being passed disappear form the sender's address space and appear in the receiver's address space. One interesting aspect of the Mach is that the microkernel itself is structured internally as a task with threads, and most of the Mach system services are implemented as IPCs to the kernel rather than as direct system calls. In order to support I/O and device access, the microkernel provides access to I/O resources such as memory-mapped devices, I/O ports and direct memory access (DMA) channels, as well as the ability to reflect interrupts to device drivers executing in user space. Mach has services that let privileged programs acquire ownership of devices: these programs are typically device driver servers running as applications as shown in the diagram. Since the microkernel must field all interrupts (because interrupts are typically delivered into the privileged state of the computer and in order to maintain the integrity of the system), it has logic that determines whether it should process the interrupt or reflect it to a server. If the interrupt is to be reflected to an application, it must have registered with the kernel and contain code that waits for the kernel to reflect the interrupt. When the interrupt is reflected, a thread in the application begins running to process the interrupt. (5) Mach's host and processor set features provide two related sets of functions that are needed if application programs are to provide most operating system services. The host features return information about the processor complex running the system and provide certain system management functions such as time, date, and system stop and restart. The processor set features are used on multiprocessor machines to group processors in classes. These classes permit a parallel application to execute multiple threads simultaneously on different processors in the same machine so that true parallel execution occurs. The diagram also shows that Mach carefully splits its implementation into code that is completely portable from machine to machine and code that is dependent on the particular machine on which it is executing. It also segregates the code that depends on devices into the device drivers: however, the device driver code, while device-dependent, is not necessarily dependent on the processor architecture. Using multiple threads per task, Mach provides an application environment that permits the use of multiprocessors without requiring that any particular machine be a multiprocessor: on uniprocessors different threads simply run at different times. All of the support needed for multiprocessors is concentrated into the relatively small and simple microkernel. OPERATING SYSTEMS AS APPLICATION SERVERS Sitting immediately on top of the microkernel are a number of application servers that provide general purpose system services. These programs are called personality-independent servers since they depend only on the microkernel and themselves. Among the personality- independent servers are the default pager, the hardware resource manager that assigns device hardware to device drivers, the master server that loads other personality-independent servers into memory, the the device drivers and possibly a general purpose file server. The microkernel and the personality-independent servers can be packaged as a product separate from any operating system. Using the services of the microkernel and the personality-independent servers, the dominant personality provides a standard operating system such as AIX. Since the dominant personality is an application server, it is possible to have multiple servers for different personalities executing programs written to different operating systems running on the machine at the same time. However, there are some operating system services such as error message translation that are not provided in the personality-independent servers. Since it is best not to duplicate such services, the dominant personality provides them not only to its client applications but also to any other personalities running on the machine. For instance, the low-end AIX implementation on Mach currently runs as a dominant personality and supports an environment for running DOS programs as a secondary personality. (6) BENEFITS FOR DEVELOPERS AND USERS Mach represents enabling technology for system developers rather than a revolution for the end user. The modular, scalable, portable environment of Mach permits the system developer to create new system functions as applications rather than within the kernel. It lets hardware developers write device drivers as applications and makes porting the system to new platforms a relatively small and very well- understood job. But there are some real benefits to the end user. The customer now has a choice: applications written to different operating systems can execute together on the same machine permitting the use of exactly the right set of programs for a particular business or organization. Mach reduces the time it takes system developers to create new software functions or to use new hardware functions, providing the end user with better technology more rapidly than in the past. Finally, Mach permits the end user to benefit from the emergence of parallel processing and the notions of object-oriented programming by providing a system that allows applications to use parallelism and that breaks traditional operating systems up into modular pieces to which object-oriented implementation techniques can be applied.