|
NT and VMS
Most of NT's core designers had worked on and with VMS at Digital; some had worked directly with Cutler. How could these developers prevent their VMS design decisions from affecting their design and implementation of NT? Many users believe that NT's developers carried concepts from VMS to NT, but most don't know just how similar NT and VMS are at the kernel level (despite the Usenet joke that if you increment each letter in VMS you end up with WNTWindows NT).
As in UNIX and most commercial OSs, NT has two modes of execution, as Figure 2 shows. In user mode, applications execute, and OS/2, DOS, and POSIX execute and export APIs for applications to use. These components are unprivileged because NT controls them and the hardware they run on. Without NT's permission, these components cannot directly access hardware. In addition, the components and hardware cannot access each other's memory space, nor can they access the memory associated with NT's kernel. The components in user mode must call on the kernel if they want to access hardware or allocate physical or logical resources.
The kernel executes in a privileged mode: It can directly access memory and hardware. The kernel consists of several Executive subsystems, which are responsible for managing resources, including the Process Manager, the I/O Manager, the Virtual Memory Manager, the Security Reference Monitor, and a microkernel that handles scheduling and interrupts. The system dynamically loads device drivers, which are kernel components that interface NT to different peripheral devices. The hardware abstraction layer (HAL) hides the specific intricacies of an underlying CPU and motherboard from NT. NT's native API is the API that user-mode applications use to speak to the kernel. This native API is mostly undocumented, because applications are supposed to speak Win32, DOS, OS/2, POSIX, or Win16, and these respective OS environments interact with the kernel on the application's behalf.
VMS doesn't have different OS personalities, as NT does, but its kernel and Executive subsystems are clear predecessors to NT's. Digital developers wrote the VMS kernel almost entirely in VAX assembly language. To be portable across different CPU architectures, Microsoft developers wrote NT's kernel almost entirely in C. In developing NT, these designers rewrote VMS in C, cleaning up, tuning, tweaking, and adding some new functionality and capabilities as they went. This statement is in danger of trivializing their efforts; after all, the designers built a new API (i.e., Win32), a new file system (i.e., NTFS), and a new graphical interface subsystem and administrative environment while maintaining backward compatibility with DOS, OS/2, POSIX, and Win16. Nevertheless, the migration of VMS internals to NT was so thorough that within a few weeks of NT's release, Digital engineers noticed the striking similarities.
Those similarities could fill a book. In fact, you can read sections of VAX/VMS Internals and Data Structures (Digital Press) as an accurate description of NT internals simply by translating VMS terms to NT terms. Table 1 lists a few VMS terms and their NT translations. Although I won't go into detail, I will discuss some of the major similarities and differences between Windows NT 3.1 and VMS 5.0, the last version of VMS Dave Cutler and his team might have influenced. This discussion assumes you have some familiarity with OS concepts (for background information about NT's architecture, see "Windows NT Architecture, Part 1" March 1998 and "Windows NT Architecture, Part 2" April 1998).
NT's processes are virtually the same as VMS's processes (Table 2, page 118, shows a comparison of VMS and NT processes). In NT, as in VMS, the process scheduler implements 32 priority levels. The process with the highest priority is always running, and processes with equal priority are scheduled in a round-robin pattern. The system considers the 16 high-priority levels realtime or fixed priorities, because the process scheduler doesn't manipulate priority in processes the system assigns to that range. The 16 low-priority levels (except 0, which the system reserves for the idle thread that executes when nothing else can) are dynamic because the scheduler, often with the input of device drivers, bumps priorities up in reaction to various conditions, such as when the process receives input from a device. This bumping procedure is called boosting. A defining aspect of the NT and VMS schedulers is that they never lower a process' priority below the priority level the application programmed. To handle CPU hogging, in which a process burns CPU cycles without regard to other processes in the system, the scheduler boosts the priority of starved processes that haven't executed for a defined period. Both VMS 5.0 and NT 3.1 schedulers support symmetric multiprocessing (SMP), which let them execute processes simultaneously on different CPUs in order to increase applications' performance.
A major difference between NT process management and VMS process management is that NT processes contain one or more threads of execution, and NT's scheduler gives CPU time to threads, not processes. Digital didn't introduce kernel threads into VMS until version 7.0 in 1995. This addition is one of several enhancements Digital has made to VMS since NT's release that appear to be in response to NT capabilities. In turn, Microsoft added lightweight user-mode threads support to NT 4.0 in 1996, which it copied from the VMS implementation of threads.
|
Register
