1 November 2020 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
14 January 2022 in ACM Transactions on Embedded Computing Systems
Hybrid memory systems, comprised of emerging non-volatile memory (NVM) and DRAM, have been proposed to address the growing memory demand of current mobile applications. Recently emerging NVM technologies, such as phase-change memories (PCM), memristor, and 3D XPoint, have higher capacity density, minimal static power consumption and lower cost per GB. However, NVM has longer access latency and limited write endurance as opposed to DRAM. The different characteristics of distinct memory classes render a new challenge for memory system design. Ideally, pages should be placed or migrated between the two types of memories according to the data objects’ access properties. Prior system software approaches exploit the program information from OS but at the cost of high software latency incurred by related kernel processes. Hardware approaches can avoid these latencies, however, hardware’s vision is constrained to a short time window of recent memory requests, due to the limited on-chip resources. In this work, we propose OpenMem: a hardware-software cooperative approach that combines the execution time advantages of pure hardware approaches with the data object properties in a global scope. First, we built a hardware-based memory manager unit (HMMU) that can learn the short-term access patterns by online profiling, and execute data migration efficiently. Then, we built a heap memory manager for the heterogeneous memory systems that allows the programmer to directly customize each data object’s allocation to a favorable memory device within the presumed object life cycle. With the programmer’s hints guiding the data placement at allocation time, data objects with similar properties will be congregated to reduce unnecessary page migrations. We implemented the whole system on the FPGA board with embedded ARM processors. In testing under a set of benchmark applications from SPEC 2017 and PARSEC, experimental results show that OpenMem reduces 44.6% energy consumption with only a 16% performance degradation compared to the all-DRAM memory system. The amount of writes to the NVM is reduced by 14% versus the HMMU-only, extending the NVM device lifetime.
16 September 2022 in ACM Transactions on Architecture and Code Optimization
Application virtual memory footprints are growing rapidly in all systems from servers down to smartphones. To address this growing demand, system integrators are incorporating ever larger amounts of main memory, warranting rethinking of memory management. In current systems, applications produce page fault exceptions whenever they access virtual memory regions that are not backed by a physical page. As application memory footprints grow, they induce more and more minor page faults. Handling of each minor page fault can take a few thousands of CPU cycles and blocks the application till the OS kernel finds a free physical frame. These page faults can be detrimental to the performance when their frequency of occurrence is high and spread across application runtime. Specifically, lazy allocation-induced minor page faults are increasingly impacting application performance. Our evaluation of several workloads indicates an overhead due to minor page faults as high as 29% of execution time. In this article, we propose to mitigate this problem through a hardware, software co-design approach. Specifically, we first propose to parallelize portions of the kernel page allocation to run ahead of fault time in a separate thread. Then we propose the Minor Fault Offload Engine (MFOE), a per-core hardware accelerator for minor fault handling. MFOE is equipped with a pre-allocated page frame table that it uses to service a page fault. On a page fault, MFOE quickly picks a pre-allocated page frame from this table, makes an entry for it in the TLB, and updates the page table entry to satisfy the page fault. The pre-allocation frame tables are periodically refreshed by a background kernel thread, which also updates the data structures in the kernel to account for the handled page faults. We evaluate this system in the gem5 architectural simulator with a modified Linux kernel running on top of simulated hardware containing the MFOE accelerator. Our results show that MFOE improves the average critical path fault handling latency by 33× and tail critical path latency by 51×. Among the evaluated applications, we observed an improvement of runtime by an average of 6.6%.
19 June 2023 in ACM Transactions on Storage
Key–value (KV) software has proven useful to a wide variety of applications including analytics, time-series databases, and distributed file systems. To satisfy the requirements of diverse workloads, KV stores have been carefully tailored to best match the performance characteristics of underlying solid-state block devices. Emerging KV storage device is a promising technology for both simplifying the KV software stack and improving the performance of persistent storage-based applications. However, while providing fast, predictable put and get operations, existing KV storage devices do not natively support range queries that are critical to all three types of applications described above. In this article, we present KVRangeDB, a software layer that enables processing range queries for existing hash-based KV solid-state disks (KVSSDs). As an effort to adapt to the performance characteristics of emerging KVSSDs, KVRangeDB implements log-structured merge tree key index that reduces compaction I/O, merges keys when possible, and provides separate caches for indexes and values. We evaluated the KVRangeDB under a set of representative workloads, and compared its performance with two existing database solutions: a Rocksdb variant ported to work with the KVSSD, and Wisckey, a key–value database that is carefully tuned for conventional block devices. On filesystem aging workloads, KVRangeDB outperforms Wisckey by 23.7× in terms of throughput and reduce CPU usage and external write amplifications by 14.3× and 9.8×, respectively.
13 March 2025 in The Journal of Supercomputing
The rapid growth of AI/ML workloads has outpaced the capabilities of CPU-centric architectures to deliver the required data throughput and compute efficiency. This paper introduces a GPU-centric architecture leveraging GPUDirect Storage (GDS) to transfer data directly from SSDs to GPU memory, bypassing CPU bottlenecks and enabling high-throughput data paths. We propose Embedding from Storage Pipelined Network (ESPN) and its extension, ESPN-LIVE, which employ optimizations like data prefetching and on-demand embedding generation to align storage latency with GPU throughput. Experiments show ESPN reduces query latency by up to $$3.9\times$$ , cuts memory usage by up to $$16\times$$ , and improves throughput by up to 68%. ESPN-LIVE eliminates the need to store multi-vector embeddings by dynamically computing document representations, reducing storage costs by up to $$16\times$$ , and making it particularly effective for single-query systems. These results highlight the potential of SSD-GPU integration for scalable, high-performance AI/ML workloads in information retrieval and LLM applications.