arXiv subjects

Samira Khan

Publications and source records attributed to Samira Khan.

12 recordsLinked to original sources

arXiv2025

SLOs-Serve: Optimized Serving of Multi-SLO LLMs

This paper introduces SLOs-Serve, a system designed for serving multi-stage large language model (LLM) requests with application- and stage-specific service level objectives (SLOs). The key idea behind SLOs-Serve is to customize the allocation of tokens to meet these SLO requirements. SLOs-Serve uses a multi-SLO dynamic programming-based algorithm to continuously optimize token allocations under SLO constraints by exploring the full design space of chunked prefill and (optional) speculative decoding. Leveraging this resource planning algorithm, SLOs-Serve effectively supports multi-SLOs and multi-replica serving with dynamic request routing while being resilient to bursty arrivals. Our evaluation across 6 LLM application scenarios (including summarization, coding, chatbot, tool calling, and reasoning) demonstrates that SLOs-Serve improves per-GPU serving capacity by 2.2x on average compared to prior state-of-the-art systems.

cs.DC↗

arXiv2024

EdgeRAG: Online-Indexed RAG for Edge Devices

Deploying Retrieval Augmented Generation (RAG) on resource-constrained edge devices is challenging due to limited memory and processing power. In this work, we propose EdgeRAG which addresses the memory constraint by pruning embeddings within clusters and generating embeddings on-demand during retrieval. To avoid the latency of generating embeddings for large tail clusters, EdgeRAG pre-computes and stores embeddings for these clusters, while adaptively caching remaining embeddings to minimize redundant computations and further optimize latency. The result from BEIR suite shows that EdgeRAG offers significant latency reduction over the baseline IVF index, but with similar generation quality while allowing all of our evaluated datasets to fit into the memory.

cs.LG↗

arXiv2023

Efficient 2D Graph SLAM for Sparse Sensing

Simultaneous localization and mapping (SLAM) plays a vital role in mapping unknown spaces and aiding autonomous navigation. Virtually all state-of-the-art solutions today for 2D SLAM are designed for dense and accurate sensors such as laser range-finders (LiDARs). However, these sensors are not suitable for resource-limited nano robots, which become increasingly capable and ubiquitous nowadays, and these robots tend to mount economical and low-power sensors that can only provide sparse and noisy measurements. This introduces a challenging problem called SLAM with sparse sensing. This work addresses the problem by adopting the form of the state-of-the-art graph-based SLAM pipeline with a novel frontend and an improvement for loop closing in the backend, both of which are designed to work with sparse and uncertain range data. Experiments show that the maps constructed by our algorithm have superior quality compared to prior works on sparse sensing. Furthermore, our method is capable of running in real-time on a modern PC with an average processing time of 1/100th the input interval time.

cs.RO↗

arXiv2022

NearPM: A Near-Data Processing System for Storage-Class Applications

Persistent Memory (PM) technologies enable program recovery to a consistent state in a case of failure. To ensure this crash-consistent behavior, programs need to enforce persist ordering by employing mechanisms, such as logging and checkpointing, which introduce additional data movement. The emerging near-data processing (NDP) architec-tures can effectively reduce this data movement overhead. In this work we propose NearPM, a near data processor that supports accelerable primitives in crash consistent programs. Using these primitives NearPM accelerate commonly used crash consistency mechanisms logging, checkpointing, and shadow-paging. NearPM further reduces the synchronization overheads between the NDP and the CPU to guarantee persistent ordering by moving ordering handling near memory. We ensures a correct persist ordering between CPU and NDP devices, as well as among multiple NDP devices with Partitioned Persist Ordering (PPO). We prototype NearPM on an FPGA platform.1 NearPM executes data-intensive operations in crash consistency mechanisms with correct ordering guarantees while the rest of the program runs on the CPU. We evaluate nine PM workloads, where each work load supports three crash consistency mechanisms -logging, checkpointing, and shadow paging. Overall, NearPM achieves 4.3-9.8X speedup in the NDP-offloaded operations and 1.22-1.35X speedup in end-to-end execution.

cs.CE↗

arXiv2018

Flexible-Latency DRAM: Understanding and Exploiting Latency Variation in Modern DRAM Chips

This article summarizes key results of our work on experimental characterization and analysis of latency variation and latency-reliability trade-offs in modern DRAM chips, which was published in SIGMETRICS 2016, and examines the work's significance and future potential. The goal of this work is to (i) experimentally characterize and understand the latency variation across cells within a DRAM chip for these three fundamental DRAM operations, and (ii) develop new mechanisms that exploit our understanding of the latency variation to reliably improve performance. To this end, we comprehensively characterize 240 DRAM chips from three major vendors, and make six major new observations about latency variation within DRAM. Notably, we find that (i) there is large latency variation across the cells for each of the three operations; (ii) variation characteristics exhibit significant spatial locality: slower cells are clustered in certain regions of a DRAM chip; and (iii) the three fundamental operations exhibit different reliability characteristics when the latency of each operation is reduced. Based on our observations, we propose Flexible-LatencY DRAM (FLY-DRAM), a mechanism that exploits latency variation across DRAM cells within a DRAM chip to improve system performance. The key idea of FLY-DRAM is to exploit the spatial locality of slower cells within DRAM, and access the faster DRAM regions with reduced latencies for the fundamental operations. Our evaluations show that FLY-DRAM improves the performance of a wide range of applications by 13.3%, 17.6%, and 19.5%, on average, for each of the three different vendors' real DRAM chips, in a simulated 8-core system.

cs.AR↗

arXiv2018

SoftMC: Practical DRAM Characterization Using an FPGA-Based Infrastructure

This paper summarizes the SoftMC DRAM characterization infrastructure, which was published in HPCA 2017, and examines the work's significance and future potential. SoftMC (Soft Memory Controller) is the first publicly-available DRAM testing infrastructure that can flexibly and efficiently test DRAM chips in a manner accessible to both software and hardware developers. SoftMC is an FPGA-based testing platform that can control and test memory modules designed for the commonly-used DDR (Double Data Rate) interface. SoftMC has two key properties: (i) it provides flexibility to thoroughly control memory behavior or to implement a wide range of mechanisms using DDR commands; and (ii) it is easy to use as it provides a simple and intuitive high-level programming interface for users, completely hiding the low-level details of the FPGA. We demonstrate the capability, flexibility, and programming ease of SoftMC with two example use cases. First, we implement a test that characterizes the retention time of DRAM cells. Second, we show that the expected latency reduction of two recently-proposed mechanisms, which rely on accessing recently-refreshed or recently-accessed DRAM cells faster than other DRAM cells, is not observable in existing DRAM chips. Various versions of the SoftMC platform have enabled many of our other DRAM characterization studies. We discuss several other use cases of SoftMC, including the ability to characterize emerging non-volatile memory modules that obey the DDR standard. We hope that our open-source release of SoftMC fills a gap in the space of publicly-available experimental memory testing infrastructures and inspires new studies, ideas, and methodologies in memory system design.

cs.AR↗

arXiv2018

Adaptive-Latency DRAM: Reducing DRAM Latency by Exploiting Timing Margins

This paper summarizes the idea of Adaptive-Latency DRAM (AL-DRAM), which was published in HPCA 2015, and examines the work's significance and future potential. AL-DRAM is a mechanism that optimizes DRAM latency based on the DRAM module and the operating temperature, by exploiting the extra margin that is built into the DRAM timing parameters. DRAM manufacturers provide a large margin for the timing parameters as a provision against two worst-case scenarios. First, due to process variation, some outlier DRAM chips are much slower than others. Second, chips become slower at higher temperatures. The timing parameter margin ensures that the slow outlier chips operate reliably at the worst-case temperature, and hence leads to a high access latency. Using an FPGA-based DRAM testing platform, our work first characterizes the extra margin for 115 DRAM modules from three major manufacturers. The experimental results demonstrate that it is possible to reduce four of the most critical timing parameters by a minimum/maximum of 17.3%/54.8% at 55C while maintaining reliable operation. AL-DRAM uses these observations to adaptively select reliable DRAM timing parameters for each DRAM module based on the module's current operating conditions. AL-DRAM does not require any changes to the DRAM chip or its interface; it only requires multiple different timing parameters to be specified and supported by the memory controller. Our real system evaluations show that AL-DRAM improves the performance of memory-intensive workloads by an average of 14% without introducing any errors. Our characterization and proposed techniques have inspired several other works on analyzing and/or exploiting different sources of latency and performance variation within DRAM chips.

cs.AR↗

arXiv2018

Decoupling GPU Programming Models from Resource Management for Enhanced Programming Ease, Portability, and Performance

The application resource specification--a static specification of several parameters such as the number of threads and the scratchpad memory usage per thread block--forms a critical component of modern GPU programming models. This specification determines the parallelism, and hence performance, of the application during execution because the corresponding on-chip hardware resources are allocated and managed based on this specification. This tight-coupling between the software-provided resource specification and resource management in hardware leads to significant challenges in programming ease, portability, and performance. Zorua is a new resource virtualization framework, that decouples the programmer-specified resource usage of a GPU application from the actual allocation in the on-chip hardware resources. Zorua enables this decoupling by virtualizing each resource transparently to the programmer. We demonstrate that by providing the illusion of more resources than physically available via controlled and coordinated virtualization, Zorua offers several important benefits: (i) Programming Ease. Zorua eases the burden on the programmer to provide code that is tuned to efficiently utilize the physically available on-chip resources. (ii) Portability. Zorua alleviates the necessity of re-tuning an application's resource usage when porting the application across GPU generations. (iii) Performance. By dynamically allocating resources and carefully oversubscribing them when necessary, Zorua improves or retains the performance of applications that are already highly tuned to best utilize the resources.

cs.DC↗

arXiv2018

Zorua: Enhancing Programming Ease, Portability, and Performance in GPUs by Decoupling Programming Models from Resource Management

The application resource specification--a static specification of several parameters such as the number of threads and the scratchpad memory usage per thread block--forms a critical component of the existing GPU programming models. This specification determines the performance of the application during execution because the corresponding on-chip hardware resources are allocated and managed purely based on this specification. This tight coupling between the software-provided resource specification and resource management in hardware leads to significant challenges in programming ease, portability, and performance, as we demonstrate in this work. Our goal in this work is to reduce the dependence of performance on the software-provided resource specification to simultaneously alleviate the above challenges. To this end, we introduce Zorua, a new resource virtualization framework, that decouples the programmer-specified resource usage of a GPU application from the actual allocation in the on-chip hardware resources. Zorua enables this decoupling by virtualizing each resource transparently to the programmer. We demonstrate that by providing the illusion of more resources than physically available, Zorua offers several important benefits: (i) Programming Ease: Zorua eases the burden on the programmer to provide code that is tuned to efficiently utilize the physically available on-chip resources. (ii) Portability: Zorua alleviates the necessity of re-tuning an application's resource usage when porting the application across GPU generations. (iii) Performance: By dynamically allocating resources and carefully oversubscribing them when necessary, Zorua improves or retains the performance of applications that are already highly tuned to best utilize the resources. The holistic virtualization provided by Zorua has many other potential uses which we describe in this paper.

cs.DC↗

arXiv2017

Understanding and Exploiting Design-Induced Latency Variation in Modern DRAM Chips

Variation has been shown to exist across the cells within a modern DRAM chip. We empirically demonstrate a new form of variation that exists within a real DRAM chip, induced by the design and placement of different components in the DRAM chip. Our goals are to understand design-induced variation that exists in real, state-of-the-art DRAM chips, exploit it to develop low-cost mechanisms that can dynamically find and use the lowest latency at which to operate a DRAM chip reliably, and, thus, improve overall system performance while ensuring reliable system operation. To this end, we first experimentally demonstrate and analyze designed-induced variation in modern DRAM devices by testing and characterizing 96 DIMMs (768 DRAM chips). Our characterization identifies DRAM regions that are vulnerable to errors, if operated at lower latency, and finds consistency in their locations across a given DRAM chip generation, due to design-induced variation. Based on our extensive experimental analysis, we develop two mechanisms that reliably reduce DRAM latency. First, DIVA Profiling uses runtime profiling to dynamically identify the lowest DRAM latency that does not introduce failures. DIVA Profiling exploits design-induced variation and periodically profiles only the vulnerable regions to determine the lowest DRAM latency at low cost. Our second mechanism, DIVA Shuffling, shuffles data such that values stored in vulnerable regions are mapped to multiple error-correcting code (ECC) codewords. Combined together, our two mechanisms reduce read/write latency by 40.0%/60.5%, which translates to an overall system performance improvement of 14.7%/13.7%/13.8% (in 2-/4-/8-core systems) across a variety of workloads, while ensuring reliable operation.

cs.AR↗

arXiv2016

Adaptive-Latency DRAM (AL-DRAM)

This paper summarizes the idea of Adaptive-Latency DRAM (AL-DRAM), which was published in HPCA 2015. The key goal of AL-DRAM is to exploit the extra margin that is built into the DRAM timing parameters to reduce DRAM latency. The key observation is that the timing parameters are dictated by the worst-case temperatures and worst-case DRAM cells, both of which lead to small amount of charge storage and hence high access latency. One can therefore reduce latency by adapting the timing parameters to the current operating temperature and the current DIMM that is being accessed. Using an FPGA-based testing platform, our work first characterizes the extra margin for 115 DRAM modules from three major manufacturers. The experimental results demonstrate that it is possible to reduce four of the most critical timing parameters by a minimum/maximum of 17.3%/54.8% at 55C while maintaining reliable operation. AL-DRAM adaptively selects between multiple different timing parameters for each DRAM module based on its current operating condition. AL-DRAM does not require any changes to the DRAM chip or its interface; it only requires multiple different timing parameters to be specified and supported by the memory controller. Real system evaluations show that AL-DRAM improves the performance of memory-intensive workloads by an average of 14% without introducing any errors.

cs.AR↗

arXiv2015

Simultaneous Multi Layer Access: A High Bandwidth and Low Cost 3D-Stacked Memory Interface

Limited memory bandwidth is a critical bottleneck in modern systems. 3D-stacked DRAM enables higher bandwidth by leveraging wider Through-Silicon-Via (TSV) channels, but today's systems cannot fully exploit them due to the limited internal bandwidth of DRAM. DRAM reads a whole row simultaneously from the cell array to a row buffer, but can transfer only a fraction of the data from the row buffer to peripheral IO circuit, through a limited and expensive set of wires referred to as global bitlines. In presence of wider memory channels, the major bottleneck becomes the limited data transfer capacity through these global bitlines. Our goal in this work is to enable higher bandwidth in 3D-stacked DRAM without the increased cost of adding more global bitlines. We instead exploit otherwise-idle resources, such as global bitlines, already existing within the multiple DRAM layers by accessing the layers simultaneously. Our architecture, Simultaneous Multi Layer Access (SMLA), provides higher bandwidth by aggregating the internal bandwidth of multiple layers and transferring the available data at a higher IO frequency. To implement SMLA, simultaneous data transfer from multiple layers through the same IO TSVs requires coordination between layers to avoid channel conflict. We first study coordination by static partitioning, which we call Dedicated-IO, that assigns groups of TSVs to each layer. We then provide a simple, yet sophisticated mechanism, called Cascaded-IO, which enables simultaneous access to each layer by time-multiplexing the IOs. By operating at a frequency proportional to the number of layers, SMLA provides a higher bandwidth (4X for a four-layer stacked DRAM). Our evaluations show that SMLA provides significant performance improvement and energy reduction (55%/18% on average for multi-programmed workloads, respectively) over a baseline 3D-stacked DRAM with very low area overhead.

cs.AR↗