A5063512602- Advanced Data Storage Technologies
- Caching and Content Delivery
- Cloud Computing and Resource Management
- 3D IC and TSV technologies
- Parallel Computing and Optimization Techniques
- Interconnection Networks and Systems
- Network Security and Intrusion Detection
- Distributed and Parallel Computing Systems
- Optimization and Search Problems
- Semiconductor materials and devices
- Network Traffic and Congestion Control
- Photonic and Optical Devices
- Advanced Queuing Theory Analysis
- Software-Defined Networks and 5G
- Mobile Ad Hoc Networks
- Software System Performance and Reliability
- Security in Wireless Sensor Networks
- Electronic Packaging and Soldering Technologies
- IoT and Edge/Fog Computing
- Green IT and Sustainability
- Peer-to-Peer Network Technologies
- Covalent Organic Framework Applications
- Network Packet Processing and Optimization
- Cellular Automata and Applications
- Electromagnetic Compatibility and Noise Suppression
Seattle University
2023-2025
University of Washington
2021-2025
Microsoft (United States)
2021-2025
Google (United States)
2025
University of California, Los Angeles
2025
University of Illinois Urbana-Champaign
2025
Menlo School
2025
FEV (Germany)
2024-2025
Microsoft Research (United Kingdom)
2020-2024
GlobalFoundries (United States)
2016-2023
Public cloud providers seek to meet stringent performance requirements and low hardware cost. A key driver of cost is main memory. Memory pooling promises improve DRAM utilization thereby reduce costs. However, challenging under requirements. This paper proposes Pond, the first memory system that both meets goals significantly reduces Pond builds on Compute Express Link (CXL) standard for load/store access pool two insights. First, our analysis production traces shows across 8-16 sockets...
Dynamic random-access memory (DRAM) is a key driver of performance and cost in public cloud servers. At the same time, significant amount DRAM underutilized due to fragmented use across Emerging interconnects such as Compute Express Link (CXL) offers path toward improving utilization through pooling. However, design space CXL-based systems large, with questions around size, reach, topology pool. using pools require navigating complex constraints performance, virtualization, management. This...
Function-as-a-Service (FaaS) serverless computing enables a simple programming model with almost unbounded elasticity. Unfortunately, current FaaS platforms achieve this flexibility at the cost of lower performance for data-intensive applications compared to serverful deployment. The ability have computation close data is key missing feature. We introduce Palette load balancing, which offers mechanism express locality platform, through hints we term "colors". maintains nature service - users...
Recent advances in the field of reinforcement learning promise a general approach to optimize networking systems. This paper argues against recent trend for generalization by introducing case study where domain-specific modeling enables application lightweight and robust techniques.
Research on flash devices almost exclusively focuses conventional SSDs, which expose a block interface. Industry, however, has standardized and is adopting Zoned Namespaces (ZNS) offer new storage interface that dominates SSDs. Continued research SSDs thus missed opportunity to unlock step-change improvement in system performance by building ZNS We argue for an immediate complete shift discuss directions.
Compute Express Link (CXL) is widely-supported interconnect standard that promises to enable memory disaggregation in data centers. CXL allows for pooling, which can be used create a shared space across multiple servers. However, does not specify how actually build pool. Existing proposals pools are expensive, as they require switches or large multi-headed devices. In this paper, we propose new design cost-effective. We call these designs Octopus topologies. Our uses small devices made...
Cloud platforms remain underutilized despite multiple proposals to improve their utilization (e.g., disaggregation, harvesting, and oversubscription). Our characterization of the resource virtual machines (VMs) in Azure reveals that, while CPU is main resource, we need provide a solution manage all resources holistically. We also observe that many VMs exhibit complementary temporal patterns, which can be leveraged oversubscription resources. Based on these insights, propose Coach: system...
Datacenters need to reduce embodied carbon emissions, particularly for flash, which accounts 40% of in servers. However, decreasing flash’s emissions is challenging due limited write endurance, more than halves with each generation denser flash. Reducing requires extending flash lifetime, stressing its endurance even further. The legacy Logical Block-Addressable Device (LBAD) interface exacerbates the problem by forcing devices perform garbage collection, leading writes. Flash-based caches...
<div class="section abstract"><div class="htmlview paragraph">In hybrid electric vehicles (HEVs), optimizing energy management and reducing system losses are critical for enhancing overall efficiency performance. This paper presents a novel control strategy the boost converter in aimed at minimizing performance by modulating to higher voltage only when necessary. Traditional approaches often lead unnecessary consumption maintaining levels even not required. In contrast, proposed...
TTL caching models have recently regained significant research interest, largely due to their ability fit popular policies such as LRU. In this extended abstract we briefly describe our recent work on two exact methods analyze cache networks. The first method generalizes existing results for line networks under renewal requests the broad class of whereby evictions are driven by stopping times. obtained further generalized, using second method, feedforward with Markov arrival processes (MAP)...
Frequency jamming is the fiercest attack tool to disrupt wireless communication and its malicious aspects have received much attention in literature. Yet, several recent works propose turn table employ so-called friendly for benefit of a network. For example, recently proposed applications include hiding channels, injection defense, access control. This work investigates practical viability by applying it real-world To that end, we implemented reactive frame-selective jammer on consumer...
Meeting tail latency Service Level Objectives (SLOs) in shared cloud networks is both important and challenging. One primary challenge determining limits on the multi-tenancy such that SLOs are met. Doing so involves estimating latency, which difficult, especially when tenants exhibit bursty behavior as common production environments. Nevertheless, recent papers past two years (Silo, QJump, PriorityMeister) show techniques for calculating based a branch of mathematical modeling called...
Many recent caching systems aim to improve miss ratios, but there is no good sense among practitioners of how much further ratios can be improved. In other words, should the community continue working on this problem? Currently, principled answer question. practice, object sizes often vary by several orders magnitude, where computing optimal ratio (OPT) known NP-hard. The few results with variable provide very weak bounds and are impractical compute traces realistic length. We propose a new...
Caches are at the heart of latency-sensitive systems. In this paper, we identify a growing challenge for design latency-minimizing caches called delayed hits. Delayed hits occur high throughput, when multiple requests to same object queue up before an outstanding cache miss is resolved. This effect increases latencies beyond predictions traditional caching models and simulations; in fact, algorithms designed as if simply didn't exist. We show that strategies -- even so 'optimal' can fail...
In this work, a controlled thermo-compression (TC) bonding process has been developed to address problems caused by interposer and laminate warpage when assembling large three-dimensional (3D) integrated circuit (IC) die on an organic substrate (laminate). By using TC bonding, thin with through-silicon-vias (TSV) is joined top while being held flat vacuum vertical pressure. A distribution plate used mitigate during 3D assembly. unique set of parameters which enables the joining severely...
TTL cache models provide an attractive unified approximation framework for caching policies like LRU and FIFO, whose exact analysis is notoriously hard. In this paper, we advance the understanding of by explicitly considering stochastic capacity constraints. We find in particular that reducing variance occupancy instrumental to optimize hit ratio online setting. To enforce such a desired low variance, propose novel extension model rewarding popular objects with longer TTLs. An feature...
TTL caching models have recently regained significant research interest, largely due to their ability fit popular policies such as LRU. In this extended abstract we briefly describe our recent work on two exact methods analyze cache networks. The first method generalizes existing results for line networks under renewal requests the broad class of whereby evictions are driven by stopping times. obtained further generalized, using second method, feedforward with Markov arrival processes (MAP)...
Many recent caching systems aim to improve miss ratios, but there is no good sense among practitioners of how much further ratios can be improved. In other words, should the community continue working on this problem? Currently, principled answer question. practice, object sizes often vary by several orders magnitude, where computing optimal ratio (OPT) known NP-hard. The few results with variable provide very weak bounds and are impractical compute traces realistic length. We propose a new...