PTPraveen Tammana

Praveen Tammana

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Praveen Tammana has a total of 60 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Yaksha-Prashna: Bridging the Semantic Gap in eBPF Bytecode Analysis and Diagnostics

    by Animesh Singh, K. Shiv Kumar, S. VenkataKeerthy, Pragna Mamidipaka, R. V. B. R. N. Aaseesh, Sayandeep Sen, Palanivel Kodeswaran, Theophilus A. Benson, Ramakrishna Upadrasta, Praveen Tammana

    Modern cloud infrastructures increasingly rely on eBPF-based Network Functions (NFs) for critical data-plane tasks, including observability, security, and load balancing. Because developing production-grade eBPF programs requires significant expertise, operators are increasingly deploying third-party, ''off-the-shelf'' solutions distributed as opaque bytecode. While the Linux kernel's eBPF verifier guarantees safety, it remains agnostic to the functional correctness of network logic; consequently, verifier-safe bytecode can still exhibit incorrect behavior, resulting in both intra-NF functional failures and inter-NF compositional failures. To effectively manage these opaque bytecodes, network operators require mechanisms to validate packet-processing behavior and diagnose violations when they occur. This paper presents \archp, a scalable system for validating eBPF bytecode and providing actionable diagnostic feedback. \archp pairs a network-aware analyzer—which reconstructs packet-processing semantics from opaque eBPF bytecode to build a Network Context Model (NCM)—with a unified query language and engine that supports both validation assertions and diagnostic retrieval queries. By employing control-flow-guided dataflow analysis and an '' analyze once, query multiple times '' design, \archp scales to complex, production-grade NFs while avoiding the computational overheads of state-of-the-art validation methods. Evaluation on 32 real-world NFs from production sources, including Cilium, Katran, and Suricata, demonstrates that \archp accurately extracts network context in milliseconds. Compared to Klint and DRACO, \archp matches expressivity while achieving 200--1000× lower validation time.

  • Open Access

    FaultSense: Fault Localization in Large-Scale Mixture-of-Experts Model Serving Infrastructure

    by Harish S A, Vignesh S, Ashwin Kurella, Rohan Gandhi, Praveen Tammana

    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

    BEACON: Benchmarking Adaptability of Hardware-Offloaded Congestion Control Algorithms

    by Meet Dadhania, Ranjitha K, Saptarshi Samanta, Sneha Aravind, Hrushikesh J S, Abed Mohammad Kamaluddin, Satananda Burla, Hemant Singh, Praveen Tammana

    Modern AI workloads demand microsecond-scale network reaction times, forcing data centers to offload congestion control algorithms (CCAs) to hardware. Simultaneously, emerging transport standards introduce diverse congestion signals like delay, CSIG, INT, and packet trimming. Understanding hardware-offloaded CCA adaptability—the speed of rate reduction and recovery—is critical, yet existing benchmarking methods force a trade-off between fidelity and flexibility. Commercial testers lack congestion signal flexibility, while software simulators fail to capture line-rate hardware behaviors. To resolve this tension, we present BEACON, a high-fidelity evaluation framework designed to benchmark adaptability of hardware-offloaded CCA. By operating as an inline, programmable data-plane middleman, BEACON reproduces authentic hardware interactions at full line rate while maintaining software-like flexibility for emerging signals. A preliminary evaluation on a testbed with 100 Gbps RoCEv2-based RNICs and Tofino demonstrate that BEACON precisely emulates sub-μs congestion windows and captures transient reaction times with negligible overhead, laying the foundation for a comprehensive benchmarking of hardware-offloaded CCAs.

  • Open Access

    GLENFINNAN: SmartNIC-Accelerated Data Processing for Efficient Vision AI Pipelines

    by Mike Wong, Ulysses Butler, Emma Farkash, Praveen Tammana, Anirudh Sivaraman, Ravi Netravali

    Modern AI vision deployments behave like continuous dataflow systems: thousands of camera streams require repeated data processing on the CPU before any neural network can run on the GPU. In multi-model DAG pipelines, these data processing steps multiply across stages, consuming significant CPU cycles and leaving GPUs underutilized. The CPU-bound nature of these tasks limits overall throughput, increases latency, and forces costly over-provisioning. GLENFINNAN is a SmartNIC-accelerated data processing system that restructures common image transformations into a streaming, packet-local, lookup-table-driven pipeline. By exploiting pixel locality, memoizing expensive arithmetic, and maintaining only minimal per-packet state, GLENFINNAN performs key data processing operations directly on the NIC. Across six representative pipelines, including CNNs and multi-model DAGs, it reduces CPU utilization by up to 91%, lowers estimated data processing latency by 40–50%, and supports 2.1–3.4× more concurrent streams per node while consuming only 6–14% of NIC logic resources. These results show that programmable NICs can act as first-class compute stages for vision workloads, reshaping system bottlenecks and improving cluster-wide efficiency.

  • Open Access

    Designing Size-aware Cryptographic Primitives for FPGA-based Accelerators

    by Aditya Peer, Rohan Sudhir Basugade, Siddhant, Neeraj Kumar Yadav, Agamdeep Singh, Praveen Tammana, Sumit Darak, Rinku Shah

    Telecom operators and military/defense applications (e.g., UAVs) prefer to offload compute-intensive cryptographic processing to FPGA-based cryptographic accelerators over fixed-function ASICs for reconfigurability and short deployment cycles. The state-of-the-art FPGA-based cryptographic accelerators are designed either for high throughput or power efficiency. We propose FlexMesh, a programmable, dynamically reconfigurable, and size-aware framework for 5G/6G cryptographic primitives. We design variable-sized FPGA-based cryptographic cores by varying the degree of parallelism, and demonstrate that if the cryptographic core is chosen wisely, we can achieve performance without trading resource or power efficiency for certain use cases. FlexMesh dynamically orchestrates multiple asymmetric cryptographic cores (i.e., variable-sized cores) based on the observed request size distribution. To efficiently utilize the deployed cryptographic cores, we design a workload-aware, dynamically configurable load balancer and a scheduler. We test our system over realistic 5G datasets, and our results demonstrate that asymmetric cryptographic cores reduce the request processing latency and improve resource and power efficiencies, compared to the baseline symmetric cryptographic cores.

  • Open Access

    PerfMon: Performance Monitoring of Host Network Stack

    by Ranjitha K., Ankit Sharma, Malsawmsanga Sailo, Arun Siddardha, Amrit Kumar, Praveen Tammana, Pravein Govindan Kannan, Priyanka Naik

  • Open Access

    Simulating LLM training workloads for heterogeneous compute and network infrastructure

    by Sumit Kumar, Arjun Temura, Naman Sharma, Ramanjeet Singh, Meet Dadhania, Praveen Tammana, Satananda Burla, Abed Mohammad Kamaluddin, Rinku Shah

  • Efficient In-Network Traffic Classification Using Programmable Switches With AdaFlow

    by Sankalp Mittal, Praveen Tammana

    In-network ML-based traffic classification using programmable switches has enabled faster decisions and reduced the cost of the security infrastructure and management overheads. However, due to constraints on per-packet operations and limited stateful memory in the switch data plane, there is a fundamental tradeoff between traffic classification accuracy and switch memory requirements. Existing works fall short of accurately classifying traffic with diverse flow characteristics while keeping the memory footprint low. In this paper, we propose AdaFlow, a system that aims to address this gap by incorporating traffic-specific heuristics while designing the in-network classifier. We evaluate the AdaFlow prototype via simulations and also on a testbed with an Intel Barefoot Tofino switch. Compared to the state-of-the-art, AdaFlow improves accuracy up to 7% for various use-cases while keeping the memory overheads similar to or lower than those of the existing systems.


  • Open Access

    iGuard: Efficient Isolation Forest Design for Malicious Traffic Detection in Programmable Switches

    by Sankalp Mittal, Harikrishnan V., Patel Heetkumar, Praveen Tammana

    Deploying machine learning (ML) models in programmable switch data planes facilitates low latency and high throughput traffic inference at line speed. However, data planes pose significant constraints due to the limited memory and minimal support for mathematical operations and data types. As a result, the only unsupervised ML models implemented in data planes to date are Isolation Forests (iForests). However, conventional iForest models yield suboptimal malicious traffic detection performance in various traffic use cases. To address this limitation, this paper proposes iGuard, the first iForest implementation that can accurately detect malicious traffic by incorporating the "knowledge" of more powerful autoencoders. We deploy iGuard in the form of a small set of whitelist rules that could be easily installed in the switch data planes. We implement iGuard using the P4 language, and assess its performance in an experimental platform based on Intel Tofino switches. Upon evaluating iGuard on various attack traffic use cases, our model can improve accuracy up to 48.3% while maintaining a similar or lower switch memory footprint over previous approaches to implement iForest models in real-world equipment.