arXiv subjects

Kostis Kaffes

Publications and source records attributed to Kostis Kaffes.

At least 19 recordsLinked to original sources

arXiv2026

Which Model for Which Role? Cost-Efficient Pareto Recommendation for LLM Agentic Workflows

LLM agentic workflows increasingly employ different models for specialized roles. Users may prefer different trade-offs between task quality and deployment cost, sometimes also considering latency. Finding good assignments incurs a separate search cost from evaluating multiple candidates. We introduce Cost-Coupled Gittins (CC-Gittins), a multi-objective extension of single-objective cost-aware Gittins search for Pareto recommendation. It exploits the coupling between evaluation and deployment costs: assignments that cost more to deploy also cost more to evaluate. The search alternates between seeking higher quality and lower deployment cost, reusing each evaluation to inform both searches. After each batch, CC-Gittins recommends assignments that are nondominated under uncertainty-adjusted estimates of quality and deployment cost. We evaluate CC-Gittins on eight quality--cost matrices from workflows with two to five roles and up to 1,681 candidate assignments. We also evaluate the method with latency as a third objective. Across the eight benchmarks, CC-Gittins recovers a substantial fraction of the true Pareto frontier at roughly 10% of the exhaustive-evaluation cost, and recovers most of the frontier with few false positives by 30%.

cs.LG↗

arXiv2026

Do Coding Agents Understand Why Systems Crash---and Why They Don't?

Coding agents are increasingly capable of fixing real-world bugs, yet they still struggle with large systems codebases such as the Linux kernel. To understand why, we test how agents reason about execution in these codebases through crash reasoning: predicting from source code whether a given input crashes the system. A correct prediction alone, however, does not establish this ability, since it may come from merely spotting suspicious code. In this paper, we propose source interventions: we pair source trees that differ by a single commit, so that the same input crashes only one of them, and test whether agents predict crashes more often on the crashing tree. Experiments with four coding agents on SysRB, our dataset of 934 such pairs from Linux and OSS-Fuzz projects, yield three main findings. First, agents are biased toward predicting no crash: they rarely raise false alarms but miss many real crashes, and even the best agent misses over a quarter of them. Second, locating the code that determines the crash outcome is not enough: agents often misjudge its effect, especially when the crash occurs far from that code. Third, testing only on crashing code is misleading: agents that merely notice suspicious code predict crashes on both source trees, which would look like correct reasoning on the crashing tree alone. Together, these findings indicate that current coding agents cannot yet reliably reason about execution in complex systems codebases, and that paired evaluation is necessary to measure this ability.

cs.SE↗

arXiv2026

Probe-and-Resume: When and How to Evaluate Agent Execution

Test-time search can improve agent performance by exploring alternative actions and using a critic to select promising branches. We study this approach in terminal-based computer-use environments, which expose only a partial view of a large underlying state, and find that the Monte Carlo tree search (MCTS) configuration we evaluate does not significantly improve task success. We identify two sources of poor critic performance: the agent's observations omit environment state needed for evaluation, and intermediate states offer limited evidence of whether a trajectory will succeed. At intermediate states, critics rank continuations only modestly better than chance, and environment access improves this little. We study when to invoke a critic and what access to give it to evaluate a node. Our findings motivate a policy that invokes the critic only at nodes the agent believes are final, rather than during partial progress. Further, the policy gives the critic read, write, and execute access to a fork of the computer's current and initial state, allowing it to probe and perform potentially destructive actions to surface any hidden state the trajectory omits. We call the resulting design Probe-and-Resume: when the agent claims to be done, the critic probes a fork of its state, and a rejected agent resumes where it stopped. On Terminal-Bench 2.1, this raises success by 8.4 percentage points over MCTS with conventional critics, and approaches the success rate of the Codex CLI agent with the same underlying model at less than half its cost.

cs.AI↗

arXiv2026

StateFork: Branchable Infrastructure for Agent Exploration

AI agents improve task success by exploring multiple trajectories, but for computer-use agents each trajectory modifies external environment state. Branching from an intermediate point is correct only when restoration is observation-equivalent - future actions produce the same observations - and practical only when creating, restoring, and discarding branch states is physically efficient. We study this problem for terminal-using agents, where tasks modify files, shell context, running processes, and local services. We introduce StateFork, a logical control plane that separates exploration policies from physical state materialization, exposing sessions, commands, snapshots, restores, and cleanup over multiple execution substrates. We also build Waypoint, a checkpoint/restore substrate for terminal execution sessions that combines filesystem layering, process checkpointing, and a persistent terminal-compatible command session. Together, StateFork and Waypoint improve terminal-agent exploration by combining sample-efficient search with efficient restoration of the right execution state. On Terminal-Bench, branch-based exploration through StateFork and Waypoint improves task completion over pass@20 at the same visited-node budget, and using Waypoint achieves 26% higher task accuracy than other execution substrates while completing exploration up to 70% faster. These results show that observation-equivalent, physically efficient execution sessions are a key systems abstraction for exploratory AI agents.

cs.OS↗

arXiv2026

TuxBot: Semantic-Aware Online OS Tuning with Large Language Models

Online OS tuning can improve long-running services, but existing controllers are poorly matched to live hosts. They treat scheduler, power, memory, and I/O controls as black-box variables and optimize a scalar reward. This view ignores cross-knob policy structure, breaks down when application metrics are unavailable, and can send a running service into degraded regions that persist after the bad setting is removed. We present TuxBot, a host-side framework for steady-state OS tuning with bounded language-model guidance. TuxBot turns knob schemas, telemetry, current configuration, recent action--response history, and retrieved prior runs into a compact decision context. A fast loop proposes low-latency updates, a slower loop periodically revises the search strategy, and every proposed change passes through typed validation before reaching kernel or sysctl interfaces. This lets the controller reason about OS-control meaning and indirect performance signals while keeping model cost, latency, and authority constrained. We evaluate TuxBot on 13 live workloads from five benchmark suites while tuning up to 41 Linux parameters. Across the suite, TuxBot improves stable-phase performance by 72.5% over default settings and by 153.3% relative to the strongest non-LLM baseline. A 30-window session costs about $0.20 in model calls. With only host-level metrics, TuxBot still outperforms baselines given direct application objectives by 93.7 percentage points, while avoiding severe degraded regions reached by structure-blind exploration.

cs.OS↗

arXiv2026

Agentic Data Environments

Autonomous agents promise substantial gains in speed, scale, and labor efficiency, but their failures can impose abrupt and often irreversible costs. The central challenge for agentic automation is therefore to increase the benefits of automation while bounding the consequences of failure. While databases remain central to modern computing, agents operate over a broader data environment spanning files, APIs, applications, and system state. In this talk, I will outline early work on Agentic Data Environments -- the execution substrate in which agents operate -- that both amplify agent capabilities and enforce safety guarantees. This perspective reframes data systems from passive stores of state into active substrates for safe, reliable execution.

cs.AI↗

arXiv2026

Outrunning LLM Cutoffs: A Live Kernel Crash Resolution Benchmark for All

Repairing system crashes discovered by kernel fuzzers like Syzkaller is a critical yet underexplored challenge in software engineering. While recent works have introduced Large Language Model (LLM) based agents for Linux kernel crash-resolution, their evaluation benchmarks are usually static and thus, do not capture the evolving nature of the Linux kernel, and suffer from potential data contamination due to LLM knowledge cutoffs. To address the above problem, we present (i) Live-kBench, an evaluation framework for self-evolving benchmarks that continuously scrapes and evaluates agents on freshly discovered kernel bugs, and (ii) kEnv, an agent-agnostic standardized crash-resolution environment for kernel compilation, execution, and feedback. This design decouples agent workflows from heavy-weight execution, enabling fair and scalable comparison across diverse agent frameworks under identical conditions. To this end, we curate an inaugural dataset of 534 Linux kernel bugs and empirically demonstrate a significant performance gap, with agents achieving up to 25% higher equivalent patch rate on bugs fixed before the LLM knowledge cutoff. Using kEnv, we benchmark three state-of-the-art agents, showing that they resolve 74% of crashes on the first attempt (plausible patches); however only ~20% of generated patches closely match developer fixes. Additionally, exposing crash resolution feedback improves crash resolution rate by 29%. Live-kBench provides the community with an evaluation infrastructure for self-evolving benchmarks that is both time and attribute sensitive; complete with a public dashboard to track agent progress on Linux kernel bugs.

cs.SE↗

arXiv2026

Speculative Actions: A Lossless Framework for Faster Agentic Systems

AI agents are increasingly deployed in complex, interactive environments, yet their runtime remains a major bottleneck for training, evaluation, and real-world use. Typical agent behavior unfolds sequentially, with each action requiring an API call that can incur substantial latency. For example, a game of chess between two state-of-the-art agents can take hours. We introduce Speculative Actions, a lossless acceleration framework for general agentic systems. Inspired by speculative execution in microprocessors and speculative decoding in LLM inference, our method uses faster models to predict likely future actions and execute them in parallel, committing only when predictions match. We evaluate speculative actions across gaming, e-commerce, and web search environments, and additionally study a lossy extension in an operating systems setting. Across domains, we achieve up to 55% next-action prediction accuracy, translating into up to 20% latency reductions. Finally, we present a cost-latency analysis that formalizes the tradeoff between speculative breadth and time savings. This analysis enables principled tuning and selective branch launching to ensure that multi-branch speculation delivers practical speedups without prohibitive cost growth.

cs.AI↗

arXiv2026

BranchBench: Aligning Database Branching with Agentic Demands

Branchable databases are evolving from developer tools to infrastructure for agentic workloads characterized by speculative mutations and non-linear state exploration. Traditional RDBMS mechanisms such as nested transactions do not provide the persistent isolation and concurrent branch management required by autonomous agents, and recent "zero-copy" designs make different trade-offs whose impact on agentic workloads remains unclear. To clarify this space, we present BranchBench, a benchmark for evaluating branchable relational DBMSes under agentic exploration. We characterize five representative workloads-agentic software engineering, failure reproduction, data curation, MCTS, and simulation-and design parameterized macrobenchmarks that execute branch-mutate-evaluate loops to reflect these workloads, along with microbenchmarks that isolate branch lifecycle costs. We evaluate state of the art systems including Neon, DoltgreSQL, Tiger Data, Xata, and PostgreSQL baselines, and find a fundamental tension: systems optimized for fast branching suffer up to 5-4000x slower reads as branches deepen, while systems optimized for fast data operations incur 25-1500x higher branch creation and switching latency. Further, no current system supports the representative workloads at scale. These results highlight the need for branch-native DBMSes designed specifically for agentic exploration.

cs.DB↗

arXiv2026

AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent

AI agents are increasingly deployed in real-world applications, including systems such as Manus, OpenClaw, and coding agents. Existing research has primarily focused on server-side efficiency, proposing methods such as caching, speculative execution, traffic scheduling, and load balancing to reduce the cost of serving agentic workloads. However, as users increasingly construct agents by composing local tools, remote APIs, and diverse models, an equally important optimization problem arises on the client side. Client-side optimization asks how developers should allocate the resources available to them, including model choice, local tools, and API budget across pipeline stages, subject to application-specific quality, cost, and latency constraints. Because these objectives depend on the task and deployment setting, they cannot be determined by server-side systems alone. We introduce AgentOpt, the first framework-agnostic Python package for client-side agent optimization. We first study model selection, a high-impact optimization lever in multi-step agent pipelines. Given a pipeline and a small evaluation set, the goal is to find the most cost-effective assignment of models to pipeline roles. This problem is consequential in practice: at matched accuracy, the cost gap between the best and worst model combinations can reach 13-32x in our experiments. To efficiently explore the exponentially growing combination space, AgentOpt implements ten search algorithms, including UCB-E, UCB-E with Low-Rank Factorization, Arm Elimination, Epsilon-LUCB, Threshold Successive Elimination, and Bayesian Optimization. Across four benchmarks, UCB-E recovers near-optimal accuracy while reducing evaluation budget by 62-76\% relative to brute-force search. Code and benchmark results available at https://agentoptimizer.github.io/agentopt/.

cs.LG↗

arXiv2026

VineLM: Trie-Based Fine-Grained Control for Agentic Workflows

Agentic workflows interleave configurable LLM stages with tool stages and often include retries or refinement loops. Existing workflow managers profile full workflow configurations offline and assign each request a static workflow-level plan that binds each configurable LLM stage to a single model, reuses that model across repeated loop iterations, and does not revisit those choices at runtime. We present VineLM, a workflow manager that enables fine-grained control by choosing the model for each stage invocation as execution unfolds under request-level objectives such as maximizing accuracy under cost or latency budgets. VineLM represents feasible executions as an annotated trie of model-choice prefixes and uses checkpointing and cascade profiling to estimate path accuracy, cost, and latency without exhaustively profiling every request on every path. At runtime, VineLM re-roots the trie after each stage invocation and replans over the remaining subtrie using the realized execution prefix and remaining latency budget. On NL2SQL and math reasoning workflows, VineLM improves the cost-latency-accuracy frontier over coarse workflow-level baselines, achieving up to 18% higher accuracy at the same per-request budget with its sparse profiling reducing offline profiling cost by 98-99.8% when compared to exhaustive profiling.

cs.DC↗

arXiv2026

Harvest: Opportunistic Peer-to-Peer GPU Caching for LLM Inference

Large Language Model (LLM) inference is increasingly constrained by GPU memory capacity rather than compute throughput, driven by growing model sizes and the linear growth of the key-value (KV) cache during autoregressive decoding. Existing approaches mitigate memory pressure by offloading model state and KV tensors to host memory, but incur substantial latency due to limited PCIe bandwidth. We present Harvest, an opportunistic GPU cache management framework that exploits high-bandwidth peer-to-peer GPU interconnects to dynamically place model weights and KV cache in unused GPU memory. Harvest treats peer GPU memory as a transient cache tier, preserving correctness while reducing data movement overhead under dynamic memory availability. We demonstrate significant throughput speedup of more than 2 times by using Harvest to accelerate the retrieval of two widely-used inference components: expert layer weights and KV cache entries.

cs.LG↗

arXiv2025

Cortex: Workflow-Aware Resource Pooling and Scheduling for Agentic Serving

We introduce Cortex, a prototype workflow-aware serving platform designed for agentic workloads. The core principle of Cortex is stage isolation: it provisions dedicated resource pools for each distinct stage of an agentic workflow. This simple yet powerful strategy mitigates inter-stage interference in compute and memory, leading to better KV cache utilization, higher throughput, and more predictable performance. By customizing resource allocation and scheduling within each distinct stage of agentic workflows, Cortex lays the groundwork for more advanced, agent-native serving paradigms, including malleable resource management, speculative execution of workflow branches, and a shared, multi-tiered cache for "agentic state."

cs.DC↗

arXiv2025

Toward Systems Foundations for Agentic Exploration

Agentic exploration, letting LLM-powered agents branch, backtrack, and search across many execution paths, demands systems support well beyond today's pass-at-k resets. Our benchmark of six snapshot/restore mechanisms shows that generic tools such as CRIU or container commits are not fast enough even in isolated testbeds, and they crumble entirely in real deployments where agents share files, sockets, and cloud APIs with other agents and human users. In this talk, we pinpoint three open fundamental challenges: fork semantics, which concerns how branches reveal or hide tentative updates; external side-effects, where fork awareness must be added to services or their calls intercepted; and native forking, which requires cloning databases and runtimes in microseconds without bulk copying.

cs.DC↗

arXiv2025

Wave: Offloading Resource Management to SmartNIC Cores

SmartNICs are increasingly deployed in datacenters to offload tasks from server CPUs, improving the efficiency and flexibility of datacenter security, networking and storage. Optimizing cloud server efficiency in this way is critically important to ensure that virtually all server resources are available to paying customers. Userspace system software, specifically, decision-making tasks performed by various operating system subsystems, is particularly well suited for execution on mid-tier SmartNIC ARM cores. To this end, we introduce Wave, a framework for offloading userspace system software to processes/agents running on the SmartNIC. Wave uses Linux userspace systems to better align system functionality with SmartNIC capabilities. It also introduces a new host-SmartNIC communication API that enables offloading of even $μ$s-scale system software. To evaluate Wave, we offloaded preexisting userspace system software including kernel thread scheduling, memory management, and an RPC stack to SmartNIC ARM cores, which showed a performance degradation of 1.1%-7.4% in an apples-to-apples comparison with on-host implementations. Wave recovered host resources consumed by on-host system software for memory management (saving 16 host cores), RPCs (saving 8 host cores), and virtual machines (an 11.2% performance improvement). Wave highlights the potential for rethinking system software placement in modern datacenters, unlocking new opportunities for efficiency and scalability.

cs.OS↗

arXiv2025

Fast Userspace Networking for the Rest of Us

After a decade of research in userspace network stacks, why do new solutions remain inaccessible to most developers? We argue that this is because they ignored (1) the hardware constraints of public cloud NICs (vNICs) and (2) the flexibility required by applications. Concerning the former, state-of-the-art proposals rely on specific NIC features (e.g., flow steering, deep buffers) that are not broadly available in vNICs. As for the latter, most of these stacks enforce a restrictive execution model that does not align well with cloud application requirements. We propose a new userspace network stack, Machnet, built for public cloud VMs. Central to Machnet is a new ''Least Common Denominator'' model, a conceptual NIC with a minimal feature set supported by all kernel-bypass vNICs. The challenge is to build a new solution with performance comparable to existing stacks while relying only on basic features (e.g., no flow steering, no RSS reconfiguration). Machnet uses a microkernel design to provide higher flexibility in application execution compared to a library OS design; we show that microkernels' inter-process communication overhead is negligible on large cloud networks.

cs.NI↗

arXiv2024

Shabari: Delayed Decision-Making for Faster and Efficient Serverless Functions

Serverless computing relieves developers from the burden of resource management, thus providing ease-of-use to the users and the opportunity to optimize resource utilization for the providers. However, today's serverless systems lack performance guarantees for function invocations, thus limiting support for performance-critical applications: we observed severe performance variability (up to 6x). Providers lack visibility into user functions and hence find it challenging to right-size them: we observed heavy resource underutilization (up to 80%). To understand the causes behind the performance variability and underutilization, we conducted a measurement study of commonly deployed serverless functions and learned that the function performance and resource utilization depend crucially on function semantics and inputs. Our key insight is to delay making resource allocation decisions until after the function inputs are available. We introduce Shabari, a resource management framework for serverless systems that makes decisions as late as possible to right-size each invocation to meet functions' performance objectives (SLOs) and improve resource utilization. Shabari uses an online learning agent to right-size each function invocation based on the features of the function input and makes cold-start-aware scheduling decisions. For a range of serverless functions and inputs, Shabari reduces SLO violations by 11-73% while not wasting any vCPUs and reducing wasted memory by 64-94% in the median case, compared to state-of-the-art systems, including Aquatope, Parrotfish, and Cypress.

cs.DC↗

arXiv2023

BPF-oF: Storage Function Pushdown Over the Network

Storage disaggregation, wherein storage is accessed over the network, is popular because it allows applications to independently scale storage capacity and bandwidth based on dynamic application demand. However, the added network processing introduced by disaggregation can consume significant CPU resources. In many storage systems, logical storage operations (e.g., lookups, aggregations) involve a series of simple but dependent I/O access patterns. Therefore, one way to reduce the network processing overhead is to execute dependent series of I/O accesses at the remote storage server, reducing the back-and-forth communication between the storage layer and the application. We refer to this approach as \emph{remote-storage pushdown}. We present BPF-oF, a new remote-storage pushdown protocol built on top of NVMe-oF, which enables applications to safely push custom eBPF storage functions to a remote storage server. The main challenge in integrating BPF-oF with storage systems is preserving the benefits of their client-based in-memory caches. We address this challenge by designing novel caching techniques for storage pushdown, including splitting queries into separate in-memory and remote-storage phases and periodically refreshing the client cache with sampled accesses from the remote storage device. We demonstrate the utility of BPF-oF by integrating it with three storage systems, including RocksDB, a popular persistent key-value store that has no existing storage pushdown capability. We show BPF-oF provides significant speedups in all three systems when accessed over the network, for example improving RocksDB's throughput by up to 2.8$\times$ and tail latency by up to 2.6$\times$.

cs.OS↗