Indian Institute of Science Bangalore

Global cloud service providers handle inference workloads for Large Language Models (LLMs) that span latency-sensitive (e.g., chatbots) and insensitive (e.g., report writing) tasks, resulting in diverse and often conflicting Service Level Agreement (SLA) requirements. Managing such mixed workloads is challenging due to the complexity of the inference serving stack, which encompasses multiple models, GPU hardware, and global data centers. Existing solutions often silo such fast and slow tasks onto separate GPU resource pools with different SLAs, but this leads to significant under-utilization of expensive accelerators due to load mismatch. In this article, we characterize the LLM serving workloads at Microsoft Office 365, one of the largest users of LLMs within Microsoft Azure cloud with over 10 million requests per day, and highlight key observations across workloads in different data center regions and across time. This is one of the first such public studies of Internet-scale LLM workloads. We use these insights to propose SageServe , a comprehensive LLM serving framework that dynamically adapts to workload demands using multi-timescale control knobs. It combines short-term request routing to data centers with long-term scaling of GPU VMs and model placement with higher lead times, and co-optimizes the routing and resource allocation problem using a traffic forecast model and an Integer Linear Programming (ILP) solution. We evaluate SageServe through real runs and realistic simulations on 10 million production requests across three regions and four open-source models. We achieve up to 25% savings in GPU-hours compared to the current baseline deployment and reduce GPU-hour wastage due to inefficient auto-scaling by 80%, resulting in a potential monthly cost savings of up to $2.5 million, while maintaining tail latency and meeting SLAs. The workload traces, our simulator harness and the SageServe scheduler are available at https://github.com/shashwatj07/SageServe.

Droplets, which are ubiquitous in nature, are formed through intriguing processes, and one such route is air-assisted atomization or aerobreakup. This review focuses on secondary atomization, particularly the breakup of an individual droplet subjected to high-speed flows. This process involves complex interfacial dynamics with multiscale deformations, ranging from global flattening to local unstable waves. The deformations occur at progressively smaller scales while interacting with the surrounding gas phase, forming a nonlinear cascade. Each local undulation serves as a precursor to a self-similar evolution or subsecondary breakup process that ends with a ligament-mediated mechanism. In practical scenarios, droplets often encounter nonuniform, unsteady, impulsive, or compressible flows, like shock waves, which pose extreme conditions. The spatiotemporal scales of the nonuniformity or unsteadiness of the external flow must be comparable with the drop deformation scales at either global or local levels to influence aerobreakup that cascades across hierarchical deformation scales. The compressible effects at high Mach numbers are interestingly shown to suppress the tendency toward breakup.

The foremost environmental challenge the world is facing today is the pollution resulting from heavy metals (HMs). HMs in soil above permissible values can harm crops, disrupt the food chain, and pose serious health risks. Therefore, it's necessary to find a suitable remediation technique. This study aims to find the most efficient leaching agent for washing soil contaminated due to mining activity. Flame Atomic Absorption Spectroscopy (FASS) was used to determine the initial concentrations of HMs in the soil. The Cd, Ni, and Pb levels exceeded the WHO/FAO permissible limits by 510%, 81.68%, and 14.58%, respectively, highlighting severe contamination risks. Soil column leaching tests were conducted via four leaching agents i.e. 0.1 N EDTA, 0.1 N HCl, 0.1 N EDTA + 0.1 N HCl, and 0.1 N FeCl3, to assess the leaching efficiencies of the HMs (Fe, Mn, Zn, Cu, Cr, Cd, Ni & Pb) in the soil through experimental elution curves. The results indicate that 0.1 N FeCl3 was the most efficient leaching solution, with removal efficiencies ranging from 64.06 % to 97.35 %. The relative concentrations of heavy metals leached using four different solutions were compared using a one-way ANOVA to see if there were any statistically significant differences.

Bitumen-stabilized material (BSM) consists of aggregates, bitumen emulsion or foamed bitumen, cement, and water in specific proportions. It enhances the mechanical properties of unbound granular materials to serve as a stronger base course in the pavement structure. It is understood that the mechanical behavior of BSM evolves with time because of distinct constituent materials and their complex interactions. This study investigates the effects of bitumen emulsion content, cement, and curing duration on the volumetric and mechanical properties of BSM prepared with two different gradations as recommended by the Asphalt Academy (TG2) with nominal maximum aggregate size (NMAS) 26.5 and Austroads with NMAS 19.0 mm. Laboratory tests and statistical analysis examined the moisture loss, air voids, and indirect tensile strength (ITS) of BSM. Results indicate that combining cement and bitumen emulsion significantly improves ITS by reducing moisture loss and air voids. Being the coarser gradation, TG2 performed better with cement because of the enhanced hydration, whereas the relatively finer Austroads gradation resulted in higher ITS without cement because of better compaction and adhesion. Without cement, all the mixtures resulted in lower strength than the recommended minimum ITS of 225 kPa, highlighting cement's critical role in strength development. Further, it was established that the Michaelis-Menten (MM) reaction kinetic model effectively describes curing behavior, predicting the curing rate, theoretical maximum strength, and minimum curing time. These findings presented in this paper provide practical insights for optimizing BSM mix designs by balancing emulsion content, cement dosage, and aggregate gradation.

The microstructure evolution and mechanical properties of an ultra-low carbon (0.007 wt%) Fe-Cr-Ni-Ti-Mo precipitation-hardened martensitic stainless steel were investigated after sub-zero and aging treatments. Sub-zero treatment at -73 degrees C for 8 h produced dislocation-rich lath martensite with negligible retained austenite. Aging between 482 degrees C-621 degrees C resulted in the formation of Ni3Ti precipitates and reverted austenite. The volume fraction of reverted austenite (Vf gamma) increased with aging temperature due to Ni partitioning but decreased above 593 degrees C because the austenite became thermally unstable and retransformed to martensite during cooling. Fine Ni3Ti precipitates enhanced strength and hardness, while their coarsening and (Vf gamma) increased at higher temperatures improved ductility. The sub-zero treated (ST) specimen showed the highest impact toughness, while aging progressively improved toughness with aging temperature and time. Transformation-induced plasticity (TRIP) of reverted austenite contributed to enhanced strength, ductility, and impact toughness in aged specimens. The ST condition was strengthened mainly through solid-solution and dislocation mechanisms, while precipitation hardening was the predominant strengthening mechanism after aging.