Kunal Rao NEC Labs America

Kunal Rao

Researcher

Integrated Systems

Posts

Bifröst: Peer-to-peer Load-balancing for Function Execution in Agentic AI Systems

/in /by

Agentic AI systems rely on Large Language Models (LLMs) to execute complex tasks by invoking external functions. The efficiency of these systems depends on how well function execution is managed, especially under heterogeneous and high-variance workloads, where function execution times can range from milliseconds to several seconds. Traditional load-balancing techniques, such as round-robin, least-loaded, and Peak-EWMA (used in Linkerd), struggle in such settings: round-robin ignores load imbalance, least-loaded reacts slowly to rapid workload shifts, and Peak-EWMA relies on latency tracking, which is ineffective for workloads with high execution time variability. In this paper, we introduce Bifröst, a peer-to-peer load-balancing mechanism that distributes function requests based on real-time active request count rather than latency estimates. Instead of relying on centralized load-balancers or client-side decisions, Bifröst enables function-serving pods to dynamically distribute load by comparing queue lengths and offloading requests accordingly. This avoids unnecessary overhead while ensuring better responsiveness under high-variance workloads. Our evaluation on open-vocabulary object detection, multi-modal understanding, and code generation workloads shows that Bifröst improves function completion time by up to 20% when processing 13,700 requests from 137 AI agents on a 32-node Kubernetes cluster, outperforming both OpenFaaS and OpenFaaS with Linkerd. In an AI-driven insurance claims processing workflow, Bifröst achieves up to 25% faster execution.

Read more

XPF: Agentic AI System for Business Workflow Automation

/in /by

In this paper, we propose a novel agentic AI system called XPF, which enables users to create “agents” using just natural language, where each agent is capable of executing complex, real-world business workflows in an accurate and reliable manner. XPF provides an interface to develop and iterate over the agent creation process and then deploy the agent in production when satisfactory results are produced consistently. The key components of XPF include: (a) planner, which leverages LLM to generate a step-by-step plan, which can further be edited by a human (b) compiler, which leverages LLM to compile the plan into a flow graph (c) executor, which handles distributed execution of the flow graph (using LLM, tools, RAG, etc.) on an underlying cluster and (d) verifier, which helps in verification of the output (through human generated tests or auto-generated tests using LLM). We develop five different agents using XPF and conduct experiments to evaluate one particular aspect i.e. difference in accuracy and reliability of the five agents with “human-generated” vs “auto-generated” plans. Our experiments show that we can get much more accurate and reliable response for a business workflow when step-by-step instructions (in natural language) are given by a human familiar with the workflow, rather than letting the LLM figure out the execution plan steps. In particular, we observe that “human-generated” plan almost always gives 100% accuracy whereas “auto-generated” plan almost never gives 100% accuracy. In terms of reliability, we observe through Rouge-L, Blue and Meteor scores, that the output from “human-generated” plan is much more reliable than “auto-generated” plan.

Read more

Latency-driven Execution of LLM-generated Application Code on the Computing Continuum

/in /by

Latency-critical applications demand quick responses. Ideally, detailed insights are preferable for the best decision making and response actions. However, in situations when detailed insights cannot be provided quickly, even basic information goes a long way in tackling the situation effectively. For example, in marine security application, it is critical to immediately notify as soon as an unauthorized vessel is seen. Hence, timely response may be prioritized over the response based on entire details. To address such latency-critical situations, in this paper, we propose a novel system called DiCE-EC, which leverages LLM to generate distributed code with speculative execution on Edge (fast and simple response using resource constrained hardware) and Cloud (detailed response using powerful hardware, but may be fast or slow depending on network conditions). DiCE-EC breaks down application into smaller components and executes them asynchronously across the edge and cloud computing continuum. As network conditions vary, we show through real-world marine security application, that DiCE-EC is effective in dynamically choosing detailed insights from cloud when received within latency-constraint, or falling back to simple response from edge to guarantee timely alert delivery. Without such dynamic selection of response from edge or cloud, existing systems either always provide simple responses or drop alerts. We perform real network measurements in the Gulf of Pozzuoli in Naples, Italy along accessible areas (inland and in a Ferry) and generate 1 million realistic measurements across four inaccessible regions, and demonstrate that DiCE-EC never misses an alert, while baseline misses up to ?4% alerts with real data and up to ?1% (10,000 alerts) with generated data.

Read more

LLM-based Distributed Code Generation and Cost-Efficient Execution in the Cloud

/in /by

The advancement of Generative Artificial Intelligence (AI), particularly Large Language Models (LLMs), is reshaping the software industry by automating code generation. Many LLM-driven distributed processing systems rely on serial code generation constrained by predefined libraries, limiting flexibility and adaptability. While some approaches enhance performance through parallel execution or optimize edge-cloud distributed processing for specific domains, they often overlook the cost implications of deployment, restricting scalability and economic feasibility across diverse cloud environments. This paper presents DiCE-C, a system that eliminates these constraints by starting directly from a natural language query. DiCE-C dynamically identifies available tools at runtime, programmatically refines LLM prompts, and employs a stepwise approach—first generating serial code and then transforming it into distributed code. This adaptive methodology enables efficient distributed execution without dependence on specific libraries. By leveraging high-level parallelism at the Application Programming Interface (API) level and managing API execution as services within a Kubernetes-based runtime, DiCE-C reduces idle GPU time and facilitates the use of smaller, cost-effective GPU instances. Experiments with a vision-based insurance application demonstrate that DiCE-C reduces cloud operational costs by up to 72% when using smaller GPUs (A6000 and A4000 GPU machines vs. A100 GPU machine) and by 32% when using identical GPUs (A100 GPU machines). This flexible and cost-efficient approach makes DiCE-C a scalable solution for deploying LLM-generated vision applications in cloud environments.

Read more

CAMTUNER: Adaptive Video Analytics Pipelines via Real-time Automated Camera Parameter Tuning

/in /by

In Video Analytics Pipelines (VAP), Analytics Units (AUs) such as object detection and face recognition operating on remote servers rely heavily on surveillance cameras to capture high-quality video streams to achieve high accuracy. Modern network cameras offer an array of parameters that directly influence video quality. While a few of such parameters, e.g., exposure, focus and white balance, are automatically adjusted by the camera internally, the others are not. We denote such camera parameters as non-automated (NAUTO) parameters. In this work, we first show that in a typical surveillance camera deployment, environmental condition changes can have significant adverse effect on the accuracy of insights from the AUs, but such adverse impact can potentially be mitigated by dynamically adjusting NAUTO camera parameters in response to changes in environmental conditions. Second, since most end-users lack the skill or understanding to appropriately configure these parameters and typically use a fixed parameter setting, we present CAMTUNER, to our knowledge, the first framework that dynamically adapts NAUTO camera parameters to optimize the accuracy of AUs in a VAP in response to adverse changes in environmental conditions. CAMTUNER is based on SARSA reinforcement learning and it incorporates two novel components: a light-weight analytics quality estimator and a virtual camera that drastically speed up offline RL training. Our controlled experiments and real-world VAP deployment show that compared to a VAP using the default camera setting, CAMTUNER enhances VAP accuracy by detecting 15.9% additional persons and 2.6%-4.2% additional cars (without any false positives) in a large enterprise parking lot. CAMTUNER opens up new avenues for elevating video analytics accuracy, transcending mere incremental enhancements achieved through refining deep-learning models.

Read more

DiCE-M: Distributed Code Generation and Execution for Marine Applications – An Edge-Cloud Approach

/in /by

Edge computing has emerged as a transformative technology that reduces application latency, improves cost efficiency, enhances security, and enables large-scale deployment of applications across various domains. In environmental monitoring, systems such as MegaSense[49], use low-cost sensors to gather and process real-time air quality data through edge-cloud collaboration, highlighting the critical role of edge computing in enabling scalable, efficient solutions. Similarly, marine science increasingly requires real-time processing and analysis of marine data from remote, resource-constrained environments. In this paper, we extend the power of edge computing by integrating it with Generative Artificial Intelligence(GenAI),specifically large language models (LLMs), to address challenges in marine science applications. We propose DiCE-M (Distributed Code generation and Execution for Marine applications), a robust system that uses LLM to generate distributed code for marine applications and then utilizes a runtime to efficiently execute it on an edge+cloud computing infrastructure. Specifically, DiCE-M leverages edge computing to execute lightweight AI models locally on unmanned surface vehicles(USVs)while offloading complex tasks to the cloud, thus balancing computational load and enabling realtime monitoring in marine environments. We use marine litter identification as an example application to demonstrate the utility of DiCE-M. Our results show that DiCE-M reduces latency by more than 2X when marine litter is not detected and cuts cloud computing costs by more than half compared to traditional cloud-based approaches. By selectively cropping and transmitting relevant image portions, DiCE-M further improves bandwidth efficiency, making it a reliable and cost-effective solution for deploying AI-driven applications on resource-constrained USVs in dynamic marine environments.

Read more

DiCE: Distributed Code generation and Execution

/in /by

Generative artificial intelligence (GenAI), specifically, Large Language Models (LLMs), have shown tremendous potential in automating several tasks and improving human productivity. Recent works have shown them to be quite useful in writing and summarizing text (articles, blogs, poems, stories, songs, etc.), answering questions, brainstorming ideas, and even writing code. Several LLMs have emerged specifically targeting code generation. Given a prompt, these LLMs can generate code in any desired programming language. Many tools like ChatGPT, CoPilot, CodeWhisperer, Cody, DeepSeek Coder, StarCoder, etc. are now routinely being used by software developers. However, most of the prior work in automatic code generation using LLMs is focused on obtaining “correct” and working code, and mainly runs on a single computer (serial code). In this paper, we take this to the next level, where LLMs are leveraged to generate code for execution on a distributed infrastructure. We propose a novel system called DiCE, which takes serial code as input and automatically generates distributed version of the code and efficiently executes it on a distributed setup. DiCE consists of two main components (a) LLM-based tool (Synthia) to understand dependencies in serial code and automatically generate distributed version of the code using specialized programming model and semantics, and (b) Runtime (Hermod) to understand the semantics in the distributed code and realize efficient execution on a cluster of machines (distributed infrastructure). DiCE currently focuses on visual programs synthesized by tools like ViperGPT [1] and VisReP [2] (serial code), automatically identifies higher-level task parallelism opportunities (e.g., parallel object detection), transforms the code to exploit the parallelism, and finally efficiently executes it on a cluster of machines. Through our experiments using 100 examples from the GQA dataset [3], we show that the serial codes generated by ViperGPT are successfully transformed into distributed codes which are then efficiently executed on a cluster of machines by DiCE. We note that DiCE correctly identifies opportunities for parallelism and distributes tasks on separate GPUs within the cluster. We observe an average speed-up of 2X, 2.95X, and 3.7X, and an average efficiency of 1, 0.74 and 0.48 for a cluster of 2 nodes, 4 nodes, and 8 nodes, respectively.

Read more

ECO-LLM: LLM-based Edge Cloud Optimization

/in /by

AI/ML techniques have been used to solve systems problems, but their applicability to customize solutions on-the-fly has been limited. Traditionally, any customization required manually changing the AI/ML model or modifying the code, configuration parameters, application settings, etc. This incurs too much time and effort, and is very painful. In this paper, we propose a novel technique using Generative Artificial Intelligence (GenAI) technology, wherein instructions can be provided in natural language and actual code to handle any customization is automatically generated, integrated and applied on-the-fly. Such capability is extremely powerful since it makes customization of application settings or solution techniques super easy. Specifically, we propose ECO-LLM (LLM-based Edge Cloud Optimization), which leverages Large Language Models (LLM) to dynamically adjust placement of application tasks across edge and cloud computing tiers, in response to changes in application workload, such that insights are delivered quickly with low cost of operation (systems problem). Our experiments with real-world video analytics applications i.e. face recognition, human attributes detection and license plate recognition show that ECO-LLM is able to automatically generate code on-the-fly and adapt placement of application tasks across edge and cloud computing tiers. We note that the trigger workload (to switch between edge and cloud) for ECO-LLM is exactly the same as the baseline (manual) and actual placement performed by ECO-LLM is only slightly different i.e. on average (across 2 days) only 1.45% difference in human attributes detection and face recognition, and 1.11% difference in license plate recognition. Although we tackle this specific systems problem in this paper, our proposed GenAI-based technique is applicable to solve other systems problems too.

Read more

CLAP: Cost and Latency-Aware Placement of Microservices on the Computing Continuum

/in /by

For microservices-based real-time stream processing applications, computing at the edge delivers fast responses for low workloads, but as workload increases, the response time starts to slow down due to limited compute capacity. Abundant compute capacity in the cloud delivers fast responses even for higher workloads but incurs very high cost of operation. For applications which can tolerate latencies up to a certain limit, using either of them has one or the other drawback and for different applications and edge infrastructures, it is non-trivial to decide when to use only edge resources and when to leverage cloud resources. In this paper, we propose CLAP, which dynamically understands the relationship between workload and application latency, and automatically adjusts placement of microservices across edge and cloud computing continuum, with the goal of jointly reducing latency as well as cost of running microservices based streaming applications. CLAP leverages Reinforcement Learning (RL) technique to learn the optimal placement for a given workload and based on the learnings, adjusts placement of microservices as the application workload changes. We conduct experiments with real-world video analytics applications and show that CLAP adapts placement of microservices in response to varying workloads and achieves low latency for applications in a cost-efficient manner. Particularly, we show that for two real world video analytics applications i.e. human attributes and face recognition, CLAP is able to reduce average cost (across 4 days at different locations) by 47% and 58% for human attributes detection and face recognition application, respectively, while consistently maintaining latency below the tolerable limit.

Read more

LARA: Latency-Aware Resource Allocator for Stream Processing Applications

/in /by

One of the key metrics of interest for stream processing applications is “latency”, which indicates the total time it takes for the application to process and generate insights from streaming input data. For mission-critical video analytics applications like surveillance and monitoring, it is of paramount importance to report an incident as soon as it occurs so that necessary actions can be taken right away. Stream processing applications are typically developed as a chain of microservices and are deployed on container orchestration platforms like Kubernetes. Allocation of system resources like “cpu” and “memory” to individual application microservices has direct impact on “latency”. Kubernetes does provide ways to allocate these resources e.g. through fixed resource allocation or through vertical pod autoscaler (VPA), however there is no straightforward way in Kubernetes to prioritize “latency” for an end-to end application pipeline. In this paper, we present LARA, which is specifically designed to improve “latency” of stream processing application pipelines. LARA uses a regression-based technique for resource allocation to individual microservices. We implement four real-world video analytics application pipelines i.e. license plate recognition, face recognition, human attributes detection and pose detection, and show that compared to fixed allocation, LARA is able to reduce latency by up to ? 2.8X and is consistently better than VPA. While reducing latency, LARA is also able to deliver over 2X throughput compared to fixed allocation and is almost always better than VPA.

Read more