Hars Vardhan

Wireless Data Center with Millimeter Wave Network

Data center networks are becoming increasingly popular for several computationally intensive applications such as search, gaming etc. Complex wiring issues in traditional data centers motivated us to consider wireless and a more flexible solution that also enables dynamic reconfiguration of data center networks. Recent release of the IEEE 802.15.3c wireless networking standard in the 60 GHz ISM band supports multi- Gigabit point-to-point link data rate. Such high data rate links are suitable for data center interconnects. In this paper, 60 GHz wireless technology is used to emulate data center network systems. It is shown that re-configurable data center networks can be designed using IEEE 802.15.3c technology.

60GHz wireless links in data center networks

A novel approach for data center networking is described in this manuscript. It is proposed to replace wires in data centers with high data-rate, point-to-point, line of sight wireless links in the 60GHz frequency band. Beam forming using phased-array transceivers is employed to create such links between node pairs. The simulation results performed on the designed transceiver have shown that a high beam rotation angle, small phase rotation step and low phase rotation latency are achievable. Instead of arranging server racks in rows, a polygonal arrangement of server racks has been proposed where the line of sight wireless links are formed in the enclosed space between racks. Data center topologies are emulated with the polygonal prism arrangement. A large volume of wires removed from data center enclosure would not only reduce the total infrastructure cost but also enhance the cooling efficiency.

Finding a simple path with multiple must-include nodes

This paper presents an algorithm to find a simple path in the given network with multiple must-include nodes in the path. The problem of finding a path with must-include node(s) can be easily found in some special cases. However, in general, including multiple nodes in the simple path has been shown to be NP-Complete. This problem may arise in network areas such as forcing the route to go through particular nodes, which have wavelength converter (optical), have monitoring provision (telecom), have gateway functions (in OSPF) or are base stations (in MANET). In this paper, a heuristic algorithm is described that follows divide and conquer approach, by dividing the problem in two subproblems. It is shown that the algorithm does not grow exponentially in this application and initial re-ordering of the given sequence of must-include nodes can improve the result. The experimental results demonstrate that the algorithm successfully computes near optimal path in reasonable time.

Continuous Active Phase Shifter Design and Analysis for Millimeter-Wave Circuits

This brief presents a continuous active phase shifter for millimeter-wave phased-array transceivers. Having an all-pass network frequency response, the proposed phase shifter provides continuous insertion phase control and nearly constant insertion loss regardless of the insertion phase variation. A complete analysis for the phase shifter is provided along with circuit measurement results. The phase shifter is fabricated in a 0.13-μm commercial CMOS technology. A phase tuning range larger than 90 ° and -7 dB of insertion loss are achieved for the phase shifter.

The PlaNet-PTN Module: a Single Layer Design Tool for Packet Transport Network

PlaNet is a multilayer network planning tool developed at the University of Texas at Dallas. This paper illustrates some of the features of PlaNet-PTN, one of the modules available in the PlaNet tool. PlaNet-PTN can be used to design and plan a single layer packet transport network (PTN). Quality of protection, routing constraints, minimization of the network equipment cost, and user’s desired run time of the tool are just some examples of the features available in PlaNet. As shown in the paper, the PlaNet-PTN planning module is able to provide, among others, optimization of Label Switched Path (LSP) routes, link capacity placement, node and link equipment configuration.

The PlaNet-OTN module: A double layer design tool for optical transport networks

PlaNet is a multilayer network planning tool designed and developed at the University of Texas at Dallas. This paper illustrates some of the features of PlaNet-OTN, one of the modules available in the PlaNet tool. PlaNet-OTN can be used to design and plan an optical transport network (OTN), which is comprised of two layers: wavelength division multiplexing (WDM) layer, which deals with wavelength allocation and routing of WDM services, and optical transport network (OTN) layer, which deals with optical data unit (ODU) equipment provisioning and routing of ODU services. Features of the PlaNet-OTN module include: multiple protection schemes and routing constraints for both WDM and ODU services, network equipment cost minimization, load balancing of traffic, and user-controlled run time of the optimization process. As shown in this paper, the PlaNet-WDM and OTN layers and optimizing equipment usage, hence reducing the cost of the whole network.

Concurrency in polygonally arranged wireless Data Centers with all line-of-sight links

The use of wireless links can provide flexibility and reconfigurability to data center networks. Recent investigations show that 60GHz millimeter-wave technology can be used in wireless data center networks. The steerable millimeter-wave transceiver works as a multi-port switch where point-to-point links are established using beam-forming. Considering an interference model, specific to the wireless data center, we propose a new algorithm for k-pod fat-tree data center topology mapping. Further, we propose to group all links of the network into sets of mutually concurrent links. By assigning a time slot for all concurrent links in a set, we devise a TDMA scheduling algorithm for the wireless network. We show that the concurrent wireless transmissions could result in O(k2) number of slots that is k times better than graph based coloring approach. By increasing the concurrency, we argue that the use of 60GHz technology is more viable in building completely wireless data centers.

Towards 60GHz wireless switching interconnect

In the era of cloud computing and high performance computing clusters, interconnection networks are in high demand and have many applications including data center networks. Typically, nodes of interconnects are placed in a closed room and are connected using a set of wires through one or more high-end switches with little or no reconfigurability to the interconnect. In this paper, we present a design of wireless switching interconnects using 60GHz technology. The Gigabit, point-to-point and extremely directional links in the 60GHz band are used to replace wired connections in such a network to provide flexibility and reconfigurability. We extend our previous investigation where a high level node placement technique was discussed to arrange nodes into a closed polygon. Here we present an initial design for exact arrangement and placement of the fat-tree nodes, suitable for small to medium size wireless switching interconnects. We argue that creating a completely wireless interconnect is viable, and present mathematical equations and a proof to support the claim.

Limiting Wavelength Converter Usage in Resilient WDM Networks

A careful wavelength assignment (WA) to lambda services must be performed to reduce the total number of wavelength converters (WCs) that are required when the wavelength continuity constraint cannot be met in wavelength division multiplexing (WDM) networks. With the successful introduction of reconfigurable optical add-drop multiplexers (ROADMs), WDM networks are now growing in size, both in the number of optical nodes and the number of wavelengths supported. Fast and memory efficient WA algorithms are required to design cost effective large WDM networks. This paper presents a scalable and efficient WA heuristic algorithm aimed at reducing the total number of WCs that are required in (large) WDM networks bearing static lambda services. The WA algorithm is applied to both unprotected and (dedicated) protected lambda services. In the latter case, the wavelength continuity constraint between the working and protection path of a lambda service is taken into consideration when non-tunable optical transceivers are employed.

A wavelength sharing and assignment heuristic to minimize the number of wavelength converters in resilient WDM networks

With the successful introduction of reconfigurable optical add-drop multiplexers (ROADMs) and related technologies, WDM networks are now growing in the number of optical nodes, wavelengths, and lambda services supported. In addition, shared path protection mechanisms — whereby lambda services are allowed to share protection wavelength channels — are possible at the optical (WDM) layer. Efficient strategies must be devised to both determine the set of services that must share a common protection wavelength channel and assign wavelengths to every service. One objective of these strategies is to minimize the total number of wavelength converters (WCs), which are required every time the wavelength continuity constraint cannot be met. This paper presents a scalable and efficient heuristic, whose goal is to minimize the number of WCs in resilient WDM networks supporting static sets of shared path protection lambda services. The heuristic comprises a set of polynomial algorithms that are executed sequentially to obtain a sub-optimal solution. In small size instances of the problem, the heuristic is compared against the optimal solution obtained from ILP formulation. For large size instances — tens of thousands of lambda services and hundreds of nodes — the heuristic yields an average number of WCs that is close to be linear in the number of services, despite the fact that the wavelength sharing factor increases.