Ashwin Kumar
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Ashwin Kumar has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Mythili Vutukuru
Indian Institute of Technology Bombay, Mumbai, India
2 shared publications - Abuhujair Khan
Indian Institute of Technology, Bombay, India
1 shared publication - Ajinkya Tanksale
Indian Institute of Technology, Bombay, India
1 shared publication - Armaan Chowfin
Indian Institute of Technology, Mumbai, India
1 shared publication - Arnav Mishra
Indian Institute of Technology, Bombay, India
1 shared publication - Mohan Rajasekhar Ajjampudi
Indian Institute of Technology, Bombay, India
1 shared publication
Latest Publications
- Open Access
- 17 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 134-141
FaultSense: Fault Localization in Large-Scale Mixture-of-Experts Model Serving Infrastructure
While Mixture-of-Experts (MoE) models scale LLM serving across hundreds of GPUs, their reliance on all-to-all communication makes them susceptible to gray failures (e.g., GPU stragglers) which inflate serving latency without explicit errors, complicating fault localization. We present FaultSense, an application-layer approach that localizes faulty GPUs and communication paths in MoE model serving replicas without host instrumentation. FaultSense introduces a lightweight MoE probe model prototype alongside a two-phase hierarchical fault localization algorithm. Modeling the cluster as a GPU communication graph, it performs: (1) a global coverage that tests all GPUs and communication paths with a small set of probes to identify suspicious components, and then (2) a drill-down into suspicious components to localize faulty GPUs or communication paths. Our design reduces the number of diagnostic tests compared to individually testing all components (up to 20×) while maintaining low memory overheads.
- 17 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 134-141
Open AccessFaultSense: Fault Localization in Large-Scale Mixture-of-Experts Model Serving Infrastructure
While Mixture-of-Experts (MoE) models scale LLM serving across hundreds of GPUs, their reliance on all-to-all communication makes them susceptible to gray failures (e.g., GPU stragglers) which inflate serving latency without explicit errors, complicating fault localization. We present FaultSense, an application-layer approach that localizes faulty GPUs and communication paths in MoE model serving replicas without host instrumentation. FaultSense introduces a lightweight MoE probe model prototype alongside a two-phase hierarchical fault localization algorithm. Modeling the cluster as a GPU communication graph, it performs: (1) a global coverage that tests all GPUs and communication paths with a small set of probes to identify suspicious components, and then (2) a drill-down into suspicious components to localize faulty GPUs or communication paths. Our design reduces the number of diagnostic tests compared to individually testing all components (up to 20×) while maintaining low memory overheads.
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 2015-2029
Toasty: Speeding Up Network I/O with Cache-Warm Buffers
Modern NICs DMA packets directly to the LLC using technologies like DDIO, reducing access latencies for networking applications. Prior work has observed several performance issues with DDIO when the working set of packet buffers does not fit into LLC. For example, the leaky DMA problem arises when incoming packets evict older packets that have not yet been processed by the application from LLC, causing them to be fetched again from main memory. While using a smaller pool of packet buffers that fits in cache is an obvious solution, this may result in the NIC running out of buffers to DMA packets into when a burst of packets arrives. This paper proposes Toasty, a system that mitigates this tradeoff between high throughput and resilience to packet loss that arises when sizing the network packet buffer pool. While prior work has proposed hardware-based solutions to this problem, Toasty is a software-only solution that can be deployed on commodity NIC hardware. Toasty manages the packet buffer pool as a LIFO stack instead of a FIFO queue, and adapts the number of buffers populated into the NIC hardware RX ring based on incoming packet load and application processing rate. Together, these changes enable Toasty to recirculate a small working set of cache-warm buffers in steady state, while falling back to a larger pool of buffers during traffic bursts. We implement Toasty over the AF_XDP kernel bypass framework, and our evaluation shows that Toasty improves network throughput for a variety of network functions by up to 78% over the default buffer pool implementation of AF_XDP. We also show that Toasty matches the performance of a small buffer pool that fits in cache, while being more resilient to traffic bursts.
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 2015-2029
Open AccessToasty: Speeding Up Network I/O with Cache-Warm Buffers
Modern NICs DMA packets directly to the LLC using technologies like DDIO, reducing access latencies for networking applications. Prior work has observed several performance issues with DDIO when the working set of packet buffers does not fit into LLC. For example, the leaky DMA problem arises when incoming packets evict older packets that have not yet been processed by the application from LLC, causing them to be fetched again from main memory. While using a smaller pool of packet buffers that fits in cache is an obvious solution, this may result in the NIC running out of buffers to DMA packets into when a burst of packets arrives. This paper proposes Toasty, a system that mitigates this tradeoff between high throughput and resilience to packet loss that arises when sizing the network packet buffer pool. While prior work has proposed hardware-based solutions to this problem, Toasty is a software-only solution that can be deployed on commodity NIC hardware. Toasty manages the packet buffer pool as a LIFO stack instead of a FIFO queue, and adapts the number of buffers populated into the NIC hardware RX ring based on incoming packet load and application processing rate. Together, these changes enable Toasty to recirculate a small working set of cache-warm buffers in steady state, while falling back to a larger pool of buffers during traffic bursts. We implement Toasty over the AF_XDP kernel bypass framework, and our evaluation shows that Toasty improves network throughput for a variety of network functions by up to 78% over the default buffer pool implementation of AF_XDP. We also show that Toasty matches the performance of a small buffer pool that fits in cache, while being more resilient to traffic bursts.
- Open Access
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 156-162
Pyramis: Domain Specific Language for Developing Multi-tier Systems
Text-based specifications are the de-facto standard for specifying complex multi-tier systems. For example, 3GPP specifications define various interfaces, messages, and message processing at the multiple inter-connected nodes of a 5G system. These standards documents tend to be verbose, and may be ambiguous or inconsistent in places, increasing programmer effort to implement them in a general purpose language. This paper presents Pyramis, a Domain Specific Language (DSL) with suitable high-level abstractions for specifying the interfaces, messages, and processing in a multi-tier system. Pyramis allows programmers to specify multi-tier systems in a concise and precise manner, and enables easy development of software based on the specifications. We also develop a translator with Pyramis that automatically generates optimized, multi-threaded C++ code for the various components of the multi-tier system from the specification, and also generates eBPF-based measurement code for computing various performance metrics. We use Pyramis to build several components in the 5G mobile packet core. We show that the specifications written in Pyramis are 2–3 × smaller than the actual reference implementation, while the auto-generated C++ code performs on par with a hand-optimized implementation. We believe that Pyramis can eventually replace verbose text specifications like the 3GPP standards documents in telecom systems.
- 3 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 156-162
Open AccessPyramis: Domain Specific Language for Developing Multi-tier Systems
Text-based specifications are the de-facto standard for specifying complex multi-tier systems. For example, 3GPP specifications define various interfaces, messages, and message processing at the multiple inter-connected nodes of a 5G system. These standards documents tend to be verbose, and may be ambiguous or inconsistent in places, increasing programmer effort to implement them in a general purpose language. This paper presents Pyramis, a Domain Specific Language (DSL) with suitable high-level abstractions for specifying the interfaces, messages, and processing in a multi-tier system. Pyramis allows programmers to specify multi-tier systems in a concise and precise manner, and enables easy development of software based on the specifications. We also develop a translator with Pyramis that automatically generates optimized, multi-threaded C++ code for the various components of the multi-tier system from the specification, and also generates eBPF-based measurement code for computing various performance metrics. We use Pyramis to build several components in the 5G mobile packet core. We show that the specifications written in Pyramis are 2–3 × smaller than the actual reference implementation, while the auto-generated C++ code performs on par with a hand-optimized implementation. We believe that Pyramis can eventually replace verbose text specifications like the 3GPP standards documents in telecom systems.