AI coding agents are rapidly transforming software development, yet researchers do not fully understand their effectiveness in specialized domains. Previous studies examined how these agents contribute code via pull requests (PRs) across repositories from various domains, making it difficult to isolate domain-specific patterns. We analyze 497 agent-contributed PRs from 162 blockchain-focused open-source GitHub repositories in the AIDev dataset. We categorize agents into high-contribution agents and low-contribution agents groups based on their activity levels in blockchain repositories. Our analysis reveals that both groups have nearly identical PR acceptance rates (\(57.8\%\) vs. \(56.2\%\)), but high-contribution agents get their PRs merged 15 hours faster. For PR size (we divided PRs to four quartiles based on total number of codes lines changed), we find no difference between the PR acceptance rate based on their code churn rate, however, we observe an increase in PR resolution time by their size, with new features (136 increase), bug fixes (132 increase), and documentation (40 hours increase) showing the strongest sensitivity to PR size. We make our replication package publicly available [4].

AI coding agents are increasingly contributing to software development, yet their impact on mobile development has received little empirical attention. In this paper, we present the first category-level empirical study of agent-generated code in open-source mobile app projects. We analyzed PR acceptance behaviors across mobile platforms, agents, and task categories using 2,901 AI-authored pull requests (PRs) in 193 verified Android and iOS open-source GitHub repositories in the AIDev dataset. We find that Android projects have received 2x more AI-authored PRs and have achieved higher PR acceptance rate (71%) than iOS (63%), with significant agent-level variation on Android. Across task categories, PRs with routine tasks (feature, fix, and ui) achieve the highest acceptance, while structural changes like refactor and build achieve lower success and longer resolution times. Furthermore, our evolution analysis shows improvement in PR resolution time on Android through mid-2025 before it declined again. Our findings offer the first evidence-based characterization of AI agents effects on OSS mobile projects and establish empirical baselines for evaluating agent-generated contributions to design platform aware agentic systems.

With the recent boom in the cryptocurrency market, hackers have been on the lookout to find novel ways of commandeering users' machine for covert and stealthy mining operations. In an attempt to expose such under-the-hood practices, this paper explores the issue of browser cryptojacking, whereby miners are secretly deployed inside browser code without the knowledge of the user. To this end, we analyze the top 50k websites from Alexa and find a noticeable percentage of sites that are indulging in this exploitative exercise often using heavily obfuscated code. Furthermore, mining prevention plug-ins, such as NoMiner, fail to flag such cleverly concealed instances. Hence, we propose a machine learning solution based on hardware-assisted profiling of browser code in real-time. A fine-grained micro-architectural footprint allows us to classify mining applications with >99% accuracy and even flags them if the mining code has been heavily obfuscated or encrypted. We build our own browser extension and show that it outperforms other plug-ins. The proposed design has negligible overhead on the user's machine and works for all standard off-the-shelf CPUs.

In this thesis, we formulate a generalized optimization problem to minimize the electricity cost of network operation while using techniques applicable to operational networks. We applied our optimization formulation to two different networks: geo-diverse data centers and cellular networks. In case of cellular networks, using traffic traces from an large operational network, we observed that an operator can save up to 22% in electricity costs by using our proposed scheme. While the percentage amount seems modest, for an operator with 7000 cellular sites in an urban setting, this translates to savings of up to 35.36 MWh annually.

Cloud infrastructure providers and data center operators spend a major portion of their operations budget on the electric bills. We present RED-BL (Relocate Energy Demand to Better Locations), a framework for determining an optimal mapping of workload to an existing set of data centers while considering the cost of workload relocation. Within each workload mapping interval, RED-BL solution exploits the geo diversity in electricity price markets. The temporal diversity in those markets is simultaneously exploited by considering a planning window comprising several mapping intervals. Using workload traces from live Internet applications and electricity prices from the US markets, RED-BL can reduce the electric bill by as much as 81% from the case when the workload is equally distributed. Compared to a single data center deployment, an average reduction of 27% in electric bill can be achieved when RED-BL uses 10 or more data centers, a common case for most operators. When compared to existing workload relocation solutions, RED-BL achieves a further reduction of 13.63%, on average. While modest, this reduction can save millions of dollars for the operators. The cost of this saving is an inexpensive computation at the start of each planning window.

Ethernet is a low-cost, flexible and high-speed transport technology, which has traditionally seen success in local area networks and is rapidly gaining popularity in metro networks. However, its control plane, primarily based on spanning tree protocol, is not well-suited for the metro and core networks. For such networks, the IETF is evaluating a proposed framework called GELS which uses GMPLS as the control plane for Ethernet data plane. In this paper, we provide a quantitative assessment of GELS for service provider networks. In particular, we perform simulations using COST239 and COST266 networks to evaluate the performance of GELS under normal network conditions, as well as under failure conditions. Under normal network conditions, we find that the use of GELS results in placement of up to 46.4% more bandwidth when compared with native Ethernet control such as RSTP. In terms of LSP acceptance, GELS shows a 45.5% improvement over RSTP. Similarly, average link utilization using GELS is significantly better than the average link utilization when the native Ethernet control plane is used. When considering single element failures, RSTP recovers by converging to a potentially new spanning tree, which may take unacceptably long. In contrast, well-known restoration and protection mechanisms of GMPLS control plane result in much faster recovery. We And that the convergence times exhibited by GELS after single element failures are orders of magnitude better than those obtained when RSTP is used.