Search

We experimentally demonstrated an end-to-end link budget optimization over clipping in 400Gbps unamplified links, showing that the clipped MB distribution outperforms the peak-power constrained 64QAM by 1dB link budget., Comment: Submitted to European Conference on Optical Communications (ECOC) 2024

As more data-intensive tasks with large footprints are deployed in virtual machines (VMs), huge pages are widely used to eliminate the increasing address translation overhead. However, once the huge page mapping is established, all the base page regions in the huge page share a single extended page table (EPT) entry, so that the hypervisor loses awareness of accesses to base page regions. None of the state-of-the-art solutions can obtain access information at base page granularity for huge pages. We observe that this can lead to incorrect decisions by the hypervisor, such as incorrect data placement in a tiered memory system and unshared base page regions when sharing pages. This paper proposes FHPM, a fine-grained huge page management for virtualization without hardware and guest OS modification. FHPM can identify access information at base page granularity, and dynamically promote and demote pages. A key insight of FHPM is to redirect the EPT huge page directory entries (PDEs) to new companion pages so that the MMU can track access information within huge pages. Then, FHPM can promote and demote pages according to the current hot page pressure to balance address translation overhead and memory usage. At the same time, FHPM proposes a VM-friendly page splitting and collapsing mechanism to avoid extra VM-exits. In combination, FHPM minimizes the monitoring and management overhead and ensures that the hypervisor gets fine-grained VM memory accesses to make the proper decision. We apply FHPM to improve tiered memory management (FHPM-TMM) and to promote page sharing (FHPM-Share). FHPM-TMM achieves a performance improvement of up to 33% and 61% over the pure huge page and base page management. FHPM-Share can save 41% more memory than Ingens, a state-of-the-art page sharing solution, with comparable performance.

Several machine learning inspired methods for perturbation-based fiber nonlinearity (PBNLC) compensation have been presented in recent literature. We critically revisit acclaimed benefits of those over non-learned methods. Numerical results suggest that learned linear processing of perturbation triplets of PB-NLC is preferable over feedforward neural-network solutions.

The memory demand of virtual machines (VMs) is increasing, while DRAM has limited capacity and high power consumption. Non-volatile memory (NVM) is an alternative to DRAM, but it has high latency and low bandwidth. We observe that the VM with heterogeneous memory may incur up to a $1.5\times$ slowdown compared to a DRAM VM, if not managed well. However, none of the state-of-the-art heterogeneous memory management designs are customized for virtualization on a real system. In this paper, we propose HMM-V, a Heterogeneous Memory Management system for Virtualization. HMM-V automatically determines page hotness and migrates pages between DRAM and NVM to achieve performance close to the DRAM system. First, HMM-V tracks memory accesses through page table manipulation, but reduces the cost by leveraging Intel page-modification logging (PML) and a multi-level queue. Second, HMM-V quantifies the ``temperature'' of page and determines the hot set with bucket-sorting. HMM-V then efficiently migrates pages with minimal access pause and handles dirty pages with the assistance of PML. Finally, HMM-V provides pooling management to balance precious DRAM across multiple VMs to maximize utilization and overall performance. HMM-V is implemented on a real system with Intel Optane DC persistent memory. The four-VM co-running results show that HMM-V outperforms NUMA balancing and hardware management (Intel Optane memory mode) by $51\%$ and $31\%$, respectively.

Derived from the regular perturbation treatment of the nonlinear Schrodinger equation, a machine learning-based scheme to mitigate the intra-channel optical fiber nonlinearity is proposed. Referred to as the perturbation theory-aided (PA) learned digital back-propagation (LDBP), the proposed scheme constructs a deep neural network (DNN) in a way similar to the split-step Fourier method: linear and nonlinear operations alternate. Inspired by the perturbation analysis, the intra-channel cross-phase modulation term is conveniently represented by matrix operations in the DNN. The introduction of this term in each nonlinear operation considerably improves the performance, as well as enables the flexibility of PA-LDBP by adjusting the numbers of spans per step. The proposed scheme is evaluated by numerical simulations of a single carrier optical fiber communication system operating at 32 Gbaud with 64-quadrature amplitude modulation and 20*80 km transmission distance. The results show that the proposed scheme achieves approximately 3.5 dB, 1.8 dB, 1.4 dB, and 0.5 dB performance gain in terms of Q2 factor over the linear compensation, when the numbers of spans per step are 1, 2, 4, and 10, respectively. Two methods are proposed to reduce the complexity of PALDBP, i.e., pruning the number of perturbation coefficients and chromatic dispersion compensation in the frequency domain for multi-span per step cases. Investigation of the performance and complexity suggests that PA-LDBP attains improved performance gains with reduced complexity when compared to LDBP in the cases of 4 and 10 spans per step.

In this theoretical work, the DSP-perceived channel in optical coherent communications is first simplified, based on which we categorize linear MIMO equalizers into four classes according to their reference locations. The entire channel inverse can be represented by a complex conjugate-dependent system, coinciding with the widely linear equalization theory. Suboptimally removing FO dynamics, relatively static channel inverses parameterized with common device and channel parameters are presented for monitoring or calibration purposes., Comment: This work was submitted to ECOC 2021. This theoretical paper also explains the principle of the experimental demonstration in Joint Transmitter and Receiver IQ Differential Phase Calibration using a single 4x8 MIMO Equalizer, Proc. Advanced Photonics Congress 2021 (SPPCom), SpTh1D.4

We propose a fiber nonlinearity post-compensation technique using the DNN and the second-order perturbation theory. We achieve 1 dB Q-factor improvement for a 32 Gbaud PDM-64-QAM at 1200 km compared to the linear dispersion compensation.