Ulrich Drepper's research while affiliated with Red Hat and other places

... They implement a two-level cache structure that dispatches the concurrent write threads efficiently and significantly mitigates the cache block competition problem with a flexible merging scheme. Tarikul et al. [25] proposed Relational Fabric. It is a near-data vertical partitioner that allows memory or storage components to perform on-the-fly transparent data transformation. ...

... [21] OS Noise Analysis on Azalea-Unikernel [22] Demo: On-The-Fly Generation of Unikernels for Software-Defined Security in Cloud Infrastructures [23] A TOSCA-Oriented Software-Defined Security Approach for Unikernel-Based Protection Clouds [24] Unikernel-based Approach for Software-Defined Security in Cloud Infrastructures [11] A Survey on Security Isolation of Virtualization, Containers, and Unikernels [5] Cloud Cyber Security: Finding an Effective Approach with Unikernels [16] Exokernel: An Operating System Architecture for Application-Level Resource Management [25] USETL: Unikernels for the Serverless Extract Transform and Load; Why Should You Settle for Less? [26] Want More Unikernels? Inflate Them! [7] Understanding and Hardening Linux Containers [27] MirageOS Unikernel with Network Acceleration for IoT Cloud Environments [28] Azalea Unikernel IO Offload Acceleration [29] Unikernel Network Functions: A Journey Beyond the Containers [30] Unikraft and the Coming Age of Unikernels [31] fASLR: Function-Based ASLR Via TrustZone-M and MPU for Resource-Constrained IoT Systems [32] Unikernels: Library Operating Systems for the Cloud [10] NCC Group Assessing Unikernel Security [14] A Syscall-Level Binary-Compatible Unikernel [33] Rethinking the Library OS from the Top Down [8] Unikernel Linux (UKL) [34] Virtualization: A Survey on Concepts, Taxonomy And Associated Security Issues [1] Virtualization and Containerization of Application Infrastructure: A Comparison [2] Uniguard Protecting Unikernels using Intel SGX [20] Occlum: Secure and Efficient Multitasking Inside a Single Enclave of Intel SGX [35] Panoply: Low-TCB Linux Applications with SGX Enclaves [36] A Design and Verification Methodology for Secure Isolated Regions [37] Container Security: Issues, Challenges, and the Road Ahead [38] Intra-unikernel Isolation with Intel Memory Protection Keys [13] A Security Perspective on Unikernels [15] A Fresh Look at the Architecture and Performance of Contemporary Isolation Platforms [12] Unikernels as Processes [39] Accelerating Disaggregated Data Centers Using Unikernel [40] To address RQ3, Figure 1 shows how many papers in the corpus were published, per year, between 1995 and 2023. Beginning in 2014 the number of papers steadily increased to five a year in both 2018 and 2019. ...

... Overall, the VirtIO standard has become a de facto standard for virtualized I/O in the industry, and is widely adopted by hypervisors and operating systems. Its flexibility and portability make it an ideal choice for embedded systems that require virtualization capabilities [31]. ...

... More specifically, in-network processing unlocks higher application performance by reducing interand intra-node communication and bypassing MPI software layers. As new classes of devices including programmable NICs/switches [11], [12], Data Processing Units (DPUs) [13], and accelerators (FPGAs, GPUs) [14]- [16] are emerging in the datacenters [17], [18], we posit that there is an unrevealed opportunity to further improve the performance by extending in-network collective processing to a new class of complex collectives. ...

... Existing proposals (e.g., [25], [27], [29], [32], [33], [34]) aim to reduce the complexity of hardware design by providing a library of building blocks to synthesize query evaluation pipelines. However, in [25], their proposal is restricted to single query execution pipeline synthesis and does not allow the synthesis of multiple query execution pipelines, while [27], [29], [32], [34] do not support the data stream query processing model with sliding window semantics. ...

... Finally, in order to implement through software the mechanisms for isolation needed by these heterogeneous architectures, virtualization is foreseen as an interesting solution (Sohal et al. 2022;Modica et al. 2018) ...

... Worse yet, it is challenging to define complex software regulation policies that account for more than a single performance metric. This contrasts with the wide range of performance metrics exported by modern platforms at multiple levels of their complex memory hierarchy-e.g. at the level of PE (ARM 2016a; Xilinx 2024b), interconnect (ARM 2016b), and memory controller (Sohal et al. 2020;Saeed et al. 2022). Third, it forces to integrate additional system-level software components at the OS (Yun et al. 2013) or hypervisor level (Modica et al. 2018;Sohal et al. 2020), with the corresponding engineering and performance overheads. ...

... While this approach achieves near-native performance due to its tight coupling with the host kernel, it does not provide adequate isolation or allow applications to customize kernel configurations or policies. To avoid the cost of request indirection, such as context switches, unikernels [17], [18], [19] are proposed to run a container and the guest kernel in the same address space. Although this approach helps mitigate the overhead, it requires significant engineering efforts to port legacy applications to unikernels. ...

... On the contrary, in-transaction page faults are rare in parallel programs that employ fine-grained synchronization, such as many programs from HTMBench [11] in which fine-grained locks were replaced by transactions. From these results, we observe that without adequate hardware 55 support, page faults can be an important performance bottleneck in those programs that have been written from scratch keeping the TM paradigm in mind (its promise of making parallel programming easier through more coarse-grained transactions [12]). Other authors have also found page faults to be a limiting factor [13] [14] [15]. ...

... These events are recorded, analyzed, and displayed to the user for further exploration. We chose to follow the recommendation given by Ulrich Drepper [14,15], to inspect the memory subsystem effects by using a few event ratios rather than absolute values because it makes the interpretation easier. The following are brief descriptions of the event ratios we used in our low-level measurements: ...