Filters

Settings

Now showing1 - 2 of 2
  • Some of the metrics are blocked by your 
    consent settings
    Item type:Project, Access status: without files ,
    Parallel Persistency OS
    The management of memory resources has always been a core task of the operating system. However, OS designers are now facing disruptive changes on the memory side: New memory types, such as high-bandwidth memory (HBM) or non-volatile memory (NVM) have arrived and more types of processing elements, such as RDMA-capable NICs, GPGPUs, or FPGA-based accelerators, access virtual memory directly via the IO-MMU. These heterogeneous memory systems (HMSs) put not only the suitability of the classical memory hierarchy and memory access patterns at question, but also challenge the way memory is managed inside the kernel.In ParPerOS, we examine new abstractions for unified but efficient and optionally crash-consistent low-level memory management for data objects in heterogeneous memory systems that may consist of volatile, persistent, distributed and other types of main memory, accessed in parallel by a multitude of different processing elements (CPU, GPU, NIC, RDMA).
      43
  • Some of the metrics are blocked by your 
    consent settings
    Item type:Project, Access status: without files ,
    Adaptable Thread-Level Address Spaces
    Processes and address spaces are traditionally understood as closely coupled concepts.A process usually has exactly one address space for all control flows (threads) running in it, which can thereby interact efficiently via shared memory.The address space defines a (logical) memory layout and a mapping function from logical addresses to the underlying memory content (virtualization).In the ATLAS project, we are researching new operating system abstractions that make it possible to use several partially different mapping functions within the same logical address space (UNIX process) in parallel and, thereby, provide individual threads (completely transparent) a partially different view.These address-space views are, in contrast to fork(), not decoupled after their creation. They only differ in explicitly defined, "multivariant" areas, while sharing all the rest.In ATLAS, we conceptually understand addresses in the logical address space as a "universal interface" to the function and data objects they refer to; their implementation can now be overloaded or access-restricted through address-space views, comparable to implementation inheritance and interface substitutability in object-oriented languages. A different view can be applied atomically to any thread and provides this overloading without any addition indirections, which would provide, for instance, very efficient means for dynamic variability at thread level.The project aims to research the described address-space views as a variability mechanism for context-specific specialization in user processes and within the operating-system kernel.We want to answer. whether the proposed concepts can be generalized, if and how they can be implemented with sufficiently low costs, and whether the application in the operating system itself also makes it possible to meet diverging application requirements in a more targeted manner.The research is mainly conducted with constructive methods through extensions in the operating system, translators, binders and loaders.In three larger case studies from the areas of instrumentation, security and real-time, we demonstrate the broad applicability of our approach for different use cases.
      50