K42 key technologies

To achieve the above mentioned goals, we have incorporated many technologies into K42. We have written several white papers (available on our web site) describing these technologies in greater detail. This section provides an overview of the key technologies used in K42. At the beginning of this section we highlighted the ones impacting our Linux API implementation.

• Object-oriented technology has been applied to the entire system. This has been used to achieve good performance through customization, to achieve good MP performance by increasing locality, to increase maintainability by isolating modifications, and to perform autonomic functions by allowing components to be hot swapped[24].

• Much traditional kernel functionality is implemented in libraries in the application's own address space, providing a large degree of customizability and reducing overhead by avoiding crossing address space boundaries to invoke system services.

• A structure that permits machine specific features such as the PowerPC inverted page table and the MIPS software-controlled TLB to be exploited in an isolated manner without compromising portability.

• System functionality implemented in user-level servers with good performance maintained via efficient IPCs similar to L4[20].

• The use of processor-specific memory (the same virtual address on different processors maps to different physical addresses) to achieve good scalable NUMA performance. This technology, combined with avoiding global data, global code paths, and global locks, allows K42's design to scale to thousands of processors.

• A K42 specific object-oriented structure, clustered objects[12], which provide an infrastructure to implement scalable services with the degree of distribution transparent to the client. This also facilitates autonomic multiprocessor computing[4] as K42 can dynamically swap between uniprocessor and multiprocessor clustered objects.

• K42 is designed to run on 64-bit architectures and we have taken advantage of 64 bits to make performance gains by, for example, using large virtually sparse arrays rather than hash tables.

• K42 is fully preemptable and most of the kernel data is pageable.

• K42 is designed to support a simultaneous mix of time-shared, real-time, and fine-grained gang-scheduled applications.

• K42 has developed deferred object deletion [12] similar to RCU [22] allowing objects to release their locks before calling other objects. This efficient programming model is crucial for multiprocessor performance and is similar to type-safe memory [16].