This paper presents a cognitive flexible-bandwidth optical interconnect architecture for datacom networks. The proposed architecture leverages silicon photonic reconfigurable all-to-all switch fabrics interconnecting top-of-rack switches arranged in a Hyper-X-like topology with a cognitive control plane for optical reconfiguration by self-supervised learning. The proposed approach makes use of a clustering algorithm to learn the traffic patterns from historical traces. We developed a heuristic algorithm for optimizing the intra-pod connectivity graph for each identified traffic pattern. Further, to mitigate the scalability issue induced by frequent clustering operations, we parameterized the learned traffic patterns by a support vector machine classifier. The classifier is trained offline by self-labeled data to enable the classification of traffic matrices during online operations, thereby facilitating cognitive reconfiguration decision making. The simulation results show that compared with a static all-to-all interconnection, the proposed approach can improve the throughput by up to
This paper proposes a machine-learning (ML)-aided cognitive approach for effective bandwidth reconfiguration in optically interconnected datacenter/high-performance computing (HPC) systems. The proposed approach relies on a Hyper-X-like architecture augmented with flexible-bandwidth photonic interconnections at large scales using a hierarchical intra/inter-POD photonic switching layout. We first formulate the problem of the connectivity graph and routing scheme optimization as a mixed-integer linear programming model. A two-phase heuristic algorithm and a joint optimization approach are devised to solve the problem with low time complexity. Then, we propose an ML-based end-to-end performance estimator design to assist the network control plane with intelligent decision making for bandwidth reconfiguration. Numerical simulations using traffic distribution profiles extracted from HPC applications traces as well as random traffic matrices verify the accuracy performance of the ML design estimator (
- NSF-PAR ID:
- 10210534
- Publisher / Repository:
- Optical Society of America
- Date Published:
- Journal Name:
- Journal of Optical Communications and Networking
- Volume:
- 13
- Issue:
- 6
- ISSN:
- 1943-0620; JOCNBB
- Format(s):
- Medium: X Size: Article No. C10
- Size(s):
- Article No. C10
- Sponsoring Org:
- National Science Foundation
More Like this
-
while reducing the end-to-end packet latency and flow completion time by up to and , respectively. -
One of the top priorities in observational astronomy is the direct imaging and characterization of extrasolar planets (exoplanets) and planetary systems. Direct images of rocky exoplanets are of particular interest in the search for life beyond the Earth, but they tend to be rather challenging targets since they are orders-of-magnitude dimmer than their host stars and are separated by small angular distances that are comparable to the classical
diffraction limit, even for the coming generation of 30 m class telescopes. Current and planned efforts for ground-based direct imaging of exoplanets combine high-order adaptive optics (AO) with a stellar coronagraph observing at wavelengths ranging from the visible to the mid-IR. The primary barrier to achieving high contrast with current direct imaging methods is quasi-static speckles, caused largely by non-common path aberrations (NCPAs) in the coronagraph optical train. Recent work has demonstrated that millisecond imaging, which effectively “freezes” the atmosphere’s turbulent phase screens, should allow the wavefront sensor (WFS) telemetry to be used as a probe of the optical system to measure NCPAs. Starting with a realistic model of a telescope with an AO system and a stellar coronagraph, this paper provides simulations of several closely related regression models that take advantage of millisecond telemetry from the WFS and coronagraph’s science camera. The simplest regression model, called the naïve estimator, does not treat the noise and other sources of information loss in the WFS. Despite its flaws, in one of the simulations presented herein, the naïve estimator provides a useful estimate of an NCPA of radian RMS ( ), with an accuracy of radian RMS in 1 min of simulated sky time on a magnitude 8 star. The bias-corrected estimator generalizes the regression model to account for the noise and information loss in the WFS. A simulation of the bias-corrected estimator with 4 min of sky time included an NCPA ofradian RMS ( ) and an extended exoplanet scene. The joint regression of the bias-corrected estimator simultaneously achieved an NCPA estimate with an accuracy of radian RMS and an estimate of the exoplanet scene that was free of the self-subtraction artifacts typically associated with differential imaging. The contrast achieved by imaging of the exoplanet scene was at a distance of from the star and at . These contrast values are comparable to the very best on-sky results obtained from multi-wavelength observations that employ both angular differential imaging (ADI) and spectral differential imaging (SDI). This comparable performance is despite the fact that our simulations are quasi-monochromatic, which makes SDI impossible, nor do they have diurnal field rotation, which makes ADI impossible. The error covariance matrix of the joint regression shows substantial correlations in the exoplanet and NCPA estimation errors, indicating that exoplanet intensity and NCPA need to be estimated self-consistently to achieve high contrast. -
We study the relationship between the input phase delays and the output mode orders when using a pixel-array structure fed by multiple single-mode waveguides for tunable orbital-angular-momentum (OAM) beam generation. As an emitter of a free-space OAM beam, the designed structure introduces a transformation function that shapes and coherently combines multiple (e.g., four) equal-amplitude inputs, with the
th input carrying a phase delay of . The simulation results show that (1) the generated OAM order ℓ is dependent on the relative phase delay ; (2) the transformation function can be tailored by engineering the structure to support different tunable ranges (e.g., , or ); and (3) multiple independent coaxial OAM beams can be generated by simultaneously feeding the structure with multiple independent beams, such that each beam has its own value for the four inputs. Moreover, there is a trade-off between the tunable range and the mode purity, bandwidth, and crosstalk, such that the increase of the tunable range leads to (a) decreased mode purity (from 91% to 75% for ), (b) decreased 3 dB bandwidth of emission efficiency (from 285 nm for to 122 nm for ), and (c) increased crosstalk within the C-band (from to when the tunable range increases from 2 to 4). -
We present a performance analysis of compact monolithic optomechanical inertial sensors that describes their key fundamental limits and overall acceleration noise floor. Performance simulations for low-frequency gravity-sensitive inertial sensors show attainable acceleration noise floors on the order of
. Furthermore, from our performance models, we devised an optimization approach for our sensor designs, sensitivity, and bandwidth trade space. We conducted characterization measurements of these compact mechanical resonators, demonstrating -products at levels of 250 kg, which highlight their exquisite acceleration sensitivity. -
We experimentally demonstrate the utilization of adaptive optics (AO) to mitigate intra-group power coupling among linearly polarized (LP) modes in a graded-index few-mode fiber (GI FMF). Generally, in this fiber, the coupling between degenerate modes inside a modal group tends to be stronger than between modes belonging to different groups. In our approach, the coupling inside the
group can be represented by a combination of orbital-angular-momentum (OAM) modes, such that reducing power coupling in OAM set tends to indicate the capability to reduce the coupling inside the group. We employ two output OAM modes and as resultant linear combinations of degenerate and modes inside the group of a GI FMF. The power coupling is mitigated by shaping the amplitude and phase of the distorted OAM modes. Each OAM mode carries an independent 20-, 40-, or 100-Gbit/s quadrature-phase-shift-keying data stream. We measure the transmission matrix (TM) in the OAM basis within group, which is a subset of the full LP TM of the FMF-based system. An inverse TM is subsequently implemented before the receiver by a spatial light modulator to mitigate the intra-modal-group power coupling. With AO mitigation, the experimental results for and modes show, respectively, that (i) intra-modal-group crosstalk is reduced by and and (ii) near-error-free bit-error-rate performance is achieved with a penalty of and , respectively.