Publications

In this paper we present pFabric, a minimalistic datacenter transport design that provides near theoretically optimal flow completion times even at the 99th percentile for short flows, while still minimizing average flow completion time for long flows. Moreover, pFabric delivers this performance with a very simple design that is based on a key conceptual insight: datacenter transport should decouple flow scheduling from rate control. For flow scheduling, packets carry a single priority number set independently by each flow; switches have very small buffers and implement a very simple priority-based scheduling/dropping mechanism. Rate control is also correspondingly simpler; flows start at line rate and throttle back only under high and persistent packet loss. We provide theoretical intuition and show via extensive simulations that the combination of these two simple mechanisms is sufficient to provide near-optimal performance.

In mice, dietary cuprizone causes brain demyelination with subsequent spontaneous remyelination upon return to normal chow. Heavy water (²H₂O) labeling with mass spectrometric analysis can be used to measure brain de novo synthesis of several myelin components including cholesterol, phospholipids, galactocereboside (GalC) and myelin-associated proteins. 24-hydroxycholesterol (24-OHC), the major metabolite of brain cholesterol, is detected in blood and is believed to be specifically derived from CNS cholesterol metabolism. We assessed changes in syntheses of myelin components in brain and of blood sterols during cuprizone-induced experimental demyelination and remyelination, with and without thyroid hormone (T3) treatment. Mice were fed cuprizone for 4 weeks, then returned to control diet and treated with either placebo or T3 (0.005 mg/day). ²H₂O was administered for the last 14 days of cuprizone diet, and for either 6, 12 or 19 days of treatment during recovery from cuprizone, after which blood and corpus callosum (CC) samples were collected (n = 5/time point/treatment). ²H incorporation into cholesterol and 24-OHC in blood and CC, and incorporation into phospholipid (PL)-palmitate, GalC, myelin basic protein (MBP) and 2',3'-cyclic-nucleotide 3'-phosphodiesterase (CNPase) in CC were measured. Cuprizone significantly (p < 0.05) decreased syntheses of cholesterol, 24-OHC, GalC, MBP, CNPase and PL-palmitate in the CC and these effects were all reversed during recovery. T3 treatment significantly (p < 0.05) increased syntheses of cholesterol, 24-OHC and palmitate compared to placebo. 24-OHC and cholesterol turnover rates in brain and blood were nearly identical and 24-OHC rates in blood paralleled rates in CC, indicating that blood 24-OHC derives primarily from the brain and reflects oligodendrocyte function. In summary, changes in synthesis of several lipid and protein components in brain during cuprizone-induced demyelination and remyelination are measurable through stable isotope labeling. Blood 24-OHC turnover rates closely reflect flux rates of brain cholesterol in response to cuprizone and T3, which alter oligodendrocyte function. Labeling of blood 24-OHC has potential as a non-invasive marker of brain de novo cholesterol synthesis and breakdown rates in demyelinating conditions.

An entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures.

Summary form only given. A photonic crystal is a three-dimensional periodic dielectric structure possessing a frequency band or bandgap over which all electromagnetic modes, spontaneous emission, and zero point fluctuations are forbidden irrespective of propagation directions. We demonstrate the successful fabrication of these tiny 3D nanostructures with lithographically controlled operating wavelengths. We have synthesized and characterized photonic crystals, up to four successive vertical layers deep in GaAs, with varying hole diameters and lattice constants. Optical transmission spectra provide evidence for an optical scale photonic band gap.

Deep learning for clinical applications is subject to stringent performance requirements, which raises a need for large labeled datasets. However, the enormous cost of labeling medical data makes this challenging. In this paper, we build a cost-sensitive active learning system for the problem of intracranial hemorrhage detection and segmentation on head computed tomography (CT). We show that our ensemble method compares favorably with the state-of-the-art, while running faster and using less memory. Moreover, our experiments are done using a substantially larger dataset than earlier papers on this topic. Since the labeling time could vary tremendously across examples, we model the labeling time and optimize the return on investment. We validate this idea by core-set selection on our large labeled dataset and by growing it with data from the wild.

Large language models (LLMs) are increasingly integrated into many online services, yet they remain cost-prohibitive to deploy due to the requirement of expensive GPU instances. Prior work has addressed the high cost of LLM serving by improving the inference engine, but less attention has been given to selecting the most cost-efficient GPU type(s) for a specific LLM service. There is a large and growing landscape of GPU types and, within these options, higher cost does not always lead to increased performance. Instead, through a comprehensive investigation, we find that three key LLM service characteristics (request size, request rate, SLO) strongly influence GPU cost efficiency, and differing GPU types are most cost efficient for differing LLM service settings. As a result, the most cost-efficient allocation for a given service is typically a mix of heterogeneous GPU types. Based on this analysis, we introduce Mélange, a GPU allocation framework that navigates these diverse LLM service characteristics and heterogeneous GPU option space to automatically and efficiently derive the minimal-cost GPU allocation for a given LLM service. We formulate the GPU allocation task as a cost-aware bin packing problem where GPUs are bins and items are slices of the service workload. Our formulation's constraints account for a service's unique characteristics, allowing Mélange to be flexible to support diverse service settings and heterogeneity-aware to adapt the GPU allocation to a specific service. Compared to using only a single GPU type, Mélange reduces deployment costs by up to 77% in conversational settings, 33% in document-based settings, and 51% in a mixed setting.

Calorie restriction (CR) reduces the rate of cell proliferation in mitotic tissues. It has been suggested that this reduction in cell proliferation may mediate CR-induced increases in longevity. However, the mechanisms that lead to CR-induced reductions in cell proliferation rates remain unclear. To evaluate the CR-induced physiological adaptations that may mediate reductions in cell proliferation rates, we altered housing temperature and access to voluntary running wheels to determine the effects of food intake, energy expenditure, percent body fat, and body weight on proliferation rates of keratinocytes, liver cells, mammary epithelial cells, and splenic T-cells in C57BL/6 mice. We found that ∼20% CR led to a reduction in cell proliferation rates in all cell types. However, lower cell proliferation rates were not observed with reductions in 1) food intake and energy expenditure in female mice housed at 27°C, 2) percent body fat in female mice provided running wheels, or 3) body weight in male mice provided running wheels compared with ad libitum-fed controls. In contrast, reductions in insulin-like growth factor I were associated with decreased cell proliferation rates. Taken together, these data suggest that CR-induced reductions in food intake, energy expenditure, percent body fat, and body weight do not account for the reductions in global cell proliferation rates observed in CR. In addition, these data are consistent with the hypothesis that reduced cell proliferation rates could be useful as a biomarker of interventions that increase longevity.

We investigate two stripped-envelope supernovae (SNe) discovered in the nearby galaxy NGC 5806 by the (i)PTF. These SNe, designated PTF12os/SN 2012P and iPTF13bvn, exploded at a similar distance from the host-galaxy center. We classify PTF12os as a Type IIb SN based on our spectral sequence; iPTF13bvn has previously been classified as Type Ib having a likely progenitor with zero age main sequence (ZAMS) mass below ~17 solar masses. Our main objective is to constrain the explosion parameters of iPTF12os and iPTF13bvn, and to put constraints on the SN progenitors. We present comprehensive datasets on the SNe, and introduce a new reference-subtraction pipeline (FPipe) currently in use by the iPTF. We perform a detailed study of the light curves (LCs) and spectral evolution of the SNe. The bolometric LCs are modeled using the hydrodynamical code HYDE. We use nebular models and late-time spectra to constrain the ZAMS mass of the progenitors. We perform image registration of ground-based images of PTF12os to archival HST images of NGC 5806 to identify a potential progenitor candidate. Our nebular spectra of iPTF13bvn indicate a low ZAMS mass of ~12 solar masses for the progenitor. The late-time spectra of PTF12os are consistent with a ZAMS mass of ~15 solar masses. We successfully identify a progenitor candidate to PTF12os using archival HST images. This source is consistent with being a cluster of massive stars. Our hydrodynamical modeling suggests that the progenitor of PTF12os had a compact He core with a mass of 3.25 solar masses, and that 0.063 solar masses of strongly mixed 56Ni was synthesized. Spectral comparisons to the Type IIb SN 2011dh indicate that the progenitor of PTF12os was surrounded by a hydrogen envelope with a mass lower than 0.02 solar masses. We also find tentative evidence that the progenitor of iPTF13bvn could have been surrounded by a small amount of hydrogen.