Photonic-aware neural networks

Neural Computing and Applications - Photonics-based neural networks promise to outperform electronic counterparts, accelerating neural network computations while reducing power consumption and...     

Idle Protection Capacity Reuse in Multiclass Optical Networks

Optical networks carrying traffic belonging to different survivability classes must ensure not only the proper survivability differentiation but also the efficient network resource utilization. Current approaches improve network resource utilization by carrying low-class preemptable traffic along high-class spare resources, which are idle (i.e., unutilized) during failure-free conditions. In this paper, the proposed idle protection capacity reuse (IR) framework generalizes the practice of reutilizing idle spare resources to both failure-free and failure conditions. The IR framework is based on the idea of exploiting idle high-class lightpath protection resources not only for routing low-class lightpaths during the provisioning phase (i.e., provisioning-phase IR, P-IR) but also for dynamically restoring low-class lightpaths during the recovery phase (i.e., recovery-phase IR, R-IR). Both P-IR and R-IR have the potential to improve network utilization while providing multiclass lightpaths with the required survivability differentiation. The numerical evaluation shows that the utilization of R-IR in dynamic restoration results in an improved low-class lightpath survivability with respect to the utilization of stub release. Moreover, if P-IR and R-IR are alternatively exploited, then the former improves the provisioning performance, while the latter guarantees a higher survivability. In the end, if P-IR and R-IR are concurrently employed in the provisioning and in the dynamic restoration of two different low classes of lightpaths, respectively, an inherent survivability differentiation is achieved     

Label Preference Schemes for Lightpath Provisioning and Restoration in Distributed GMPLS Networks

In wavelength-routed networks based on a GMPLS control plane, the resource reservation protocol with traffic engineering extensions (RSVP-TE) allows to establish end-to-end lightpaths. The resource reservation can be blocked due to lack of available resources (forward blocking) or due to resource contentions (backward blocking). In wavelength-routed networks, the backward blocking is the predominant blocking contribution, when traffic load is low or highly-dynamic and when lightpath restoration takes place. To reduce the backward blocking, the paper proposes two label preference (LP) schemes compliant with RSVP-TE message exchanges. LP schemes provide the destination node with a label identifying the preferred wavelength to reserve. The preferred label is computed in a distributed way during the forward signaling phase, with the objective of assigning disjoint wavelengths to reservation attempts that may contend the resources. Simulation results demonstrate that, compared to other schemes, LP schemes are effective in reducing the backward blocking during both lightpath provisioning and restoration, without negatively impacting the forward blocking.     

Challenges and requirements for introducing impairment-awareness into the management and control planes of ASON/GMPLS WDM networks

The absence of electrical regenerators in transparent WDM networks significantly contributes to reduce the overall network cost. In transparent WDM networks, a proper resource allocation requires that the presence of physical impairments in routing and wavelength assignment (RWA) and lightpath provisioning be taken into account. In this article a centralized, a hybrid centralized-distributed and two distributed approaches that integrate information about most relevant physical impairments in RWA and lightpath provisioning are presented and assessed. Both centralized and hybrid approaches perform a centralized path computation at the management-plane level, utilizing physical impairment information, while the lightpath provisioning is done by the management plane or the control plane, respectively. The distributed approaches fall entirely within the scope of the ASON/GMPLS control plane. For these two approaches, we provide functional requirements, architectural functional blocks, and protocol extensions for implementing either an impairment-aware real-time RWA, or a lightpath provisioning based on impairment-aware signaling     

Trazodone versus flunitrazepam in premedication in day-care surgery

A prospective single-blind study was conducted to compare flunitrazepam vs trazodone in the premedication of patients undergoing day-case surgery for termination of pregnancy, with particular regard to the degree of preoperative sedation, intraoperative analgesia and postoperative recovery. 86 patients were randomly allocated to receive orally 45 minutes before the surgical procedure either flunitrazepam 2 mg (group F) or trazodone 50 mg (group T). In both groups anaesthesia was achieved by i.v. fentanyl 2.5 micrograms/kg and ketamina 250 micrograms/kg. Patients in group F showed a deeper degree of preoperative sedation. There were no significant differences in intraoperative analgesia and in the immediate arousal time. In the postoperative period, the incidences of emetic symptoms and dizziness were similar in both groups; the incidence of drowsiness was significantly higher in group F at 120 minutes but not at 180 minutes of observation. Psychomotor performance was assessed preoperatively two days before the surgical procedure and 60, 120 and 180 minutes after surgery, using the Toulouse-Pieron test and the reaction time to a luminous stimulus with the aid of a computerized analogic tachystoscope (Neurometer). Trazodone allowed a more rapid recovery of psychomotor performance and it can represent a valid alternative to the use of benzodiazepines in the premedication of day-case surgical patients.     

Nodal Stub-Release in All-Optical Networks

In wavelength-routed networks, the scarcity of wavelength-converters severely impacts successful connection recovery. This can be mitigated by releasing the connection's stubs (i.e. the surviving upstream and downstream span and node resources) prior to the restoration process. Releasing span resources however heavily complicates the reversion process. In this study, we present a nodal stub-release method, where only node resources (i.e. wavelength-converters) are released, while the stubs' span resources are kept occupied. Our simulation results show that the nodal stub-release method can match the performance of full stub-release (releasing both span and node resources) while keeping complexity low.     

All-optical packet-switched interconnection network based on modular photonic digital processing

Future interconnection networks will be required to achieve ultra-high bandwidth and low latency communications to cope with the increasing performance requirements of backbone routers, large data storage systems and supercomputing systems. Aiming at achieving ultra-high bandwidth communications and approaching optical time-of-flight processing latency while being robust to cascade impairments, the authors propose an all-optical packet-switched interconnection network, where not only the actual packet switching but also the packet processing is performed in the photonic domain. The authors present two modular architectures, based on the crossbar and the Batcher?Banyan topologies, capable of forwarding fixed-length packets with two classes of service. Both use photonic digital-processing subsystems built by combining a single integrable module which exploits cross gain modulation in a semiconductor optical amplifier. System level simulations on the crossbar switch controller guarantee that the control signals maintain an acceptable quality during the processing. Moreover, the Batcher? Banyan configuration is more cost-effective than the crossbar for increasing port count, while effective network performance in terms of packet loss rate can be obtained by adding just few recirculating delay lines.     

Label preference schemes in GMPLS controlled networks

The GMPLS assumption that all available labels are equal is reasonable in electronic networks but not always true in WDM optical networks where labels correspond to physical wavelengths. In this paper we present two schemes for collecting the preference for specific labels during GMPLS signaling. For this purpose a new use of the Suggested Label object is proposed, and a novel object called Suggested Vector is introduced. The approach is validated through simulations showing significant wavelength converter usage reduction in a WDM optical network     

Distributing Shared Regenerator Information in GMPLS-Controlled Translucent Networks

Sparse opto-electronic regenerators are used in optical networks to overcome optical signal quality degradations. Regenerators can be time-shared by dynamic lightpaths. However, current GMPLS control plane does not distribute regenerator information to network nodes. Three regenerator information distribution methods for GMPLS-controlled networks are proposed. The first one, Regeneration Availability during Signaling (RAS), is based on the RSVP- TE signaling protocol, while the others, Regeneration Capability Advertisement (RCA) and Regeneration Availability Advertisement (RAA), are based on the OSPF-TE routing protocol. Simulations show that RAA obtains the lowest blocking, while RCA the highest. Moreover, RAA only slightly increases the control plane load compared to RAS.