“…3 gives an example of a multi-stage network with L = 4. Multi-stage networks are the common structures in data center infrastructures like Fat-tree [15], [58], Clos [51], [53], and the direct-connect topology with spine blocks removed [11], [59].…”
Section: Models Definitionsmentioning
confidence: 99%
“…Meanwhile, the slowdown of Moore's law compared with traffic growth further increases the likelihood of overload [9], [10]. These facts pose challenges for network service providers to utilize communication bandwidth effectively to maintain network performance under overload [11]- [13]. Another source of overload is capacity reduction due to network breakdowns, including network drains during data center maintenance [11], [14], unexpected failures of nodes and links [11], [15], and cyberattacks such as denial-of-service [16], [17] Xinyu Wu and Eytan Modiano are with the Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, MA, 02139 USA.…”
We develop link rate control policies to minimize the queueing delay of packets in overloaded networks. We show that increasing link rates does not guarantee delay reduction during overload. We consider a fluid queueing model that facilitates explicit characterization of the queueing delay of packets, and establish explicit conditions on link rates that can minimize the average and maximum queueing delay in both single-hop and multi-stage (switching) networks. These min-delay conditions require maintaining an identical ratio between the ingress and egress rates of different nodes at the same layer of the network. We term the policies that follow these conditions rate-proportional policies. We further generalize the rate-proportional policies to queue-proportional policies, which minimize the queueing delay asymptotically based on the time-varying queue length while remaining agnostic of packet arrival rates. We validate that the proposed policies lead to minimum queueing delay under various network topologies and settings, compared with benchmarks including the backpressure policy that maximizes network throughput and the max-link-rate policy that fully utilizes bandwidth. We further remark that the explicit min-delay policy design in multi-stage networks facilitates co-optimization with other metrics, such as minimizing total bandwidth, balancing link utilization and node buffer usage. This demonstrates the wider utility of our main results in data center network optimization in practice.
“…3 gives an example of a multi-stage network with L = 4. Multi-stage networks are the common structures in data center infrastructures like Fat-tree [15], [58], Clos [51], [53], and the direct-connect topology with spine blocks removed [11], [59].…”
Section: Models Definitionsmentioning
confidence: 99%
“…Meanwhile, the slowdown of Moore's law compared with traffic growth further increases the likelihood of overload [9], [10]. These facts pose challenges for network service providers to utilize communication bandwidth effectively to maintain network performance under overload [11]- [13]. Another source of overload is capacity reduction due to network breakdowns, including network drains during data center maintenance [11], [14], unexpected failures of nodes and links [11], [15], and cyberattacks such as denial-of-service [16], [17] Xinyu Wu and Eytan Modiano are with the Laboratory for Information and Decision Systems, Massachusetts Institute of Technology, Cambridge, MA, 02139 USA.…”
We develop link rate control policies to minimize the queueing delay of packets in overloaded networks. We show that increasing link rates does not guarantee delay reduction during overload. We consider a fluid queueing model that facilitates explicit characterization of the queueing delay of packets, and establish explicit conditions on link rates that can minimize the average and maximum queueing delay in both single-hop and multi-stage (switching) networks. These min-delay conditions require maintaining an identical ratio between the ingress and egress rates of different nodes at the same layer of the network. We term the policies that follow these conditions rate-proportional policies. We further generalize the rate-proportional policies to queue-proportional policies, which minimize the queueing delay asymptotically based on the time-varying queue length while remaining agnostic of packet arrival rates. We validate that the proposed policies lead to minimum queueing delay under various network topologies and settings, compared with benchmarks including the backpressure policy that maximizes network throughput and the max-link-rate policy that fully utilizes bandwidth. We further remark that the explicit min-delay policy design in multi-stage networks facilitates co-optimization with other metrics, such as minimizing total bandwidth, balancing link utilization and node buffer usage. This demonstrates the wider utility of our main results in data center network optimization in practice.
“…The Si photonics platform is gaining attention as a promising solution to address the exponential growth in energy consumption within data centers resulting from the radical increase in data transmission. [1][2][3][4] A Si waveguide on a Sion-insulator wafer enables strong optical confinement thanks to the significant refractive index contrast between Si and SiO 2 , leading to the miniaturization of photonic ICs. 5) To enhance the potential of the Si photonics platform, fundamental components like lasers, photodetectors, optical modulators have undergone intensive research.…”
We examine the influence of doping profile optimization on the trade-off relationship between modulation bandwidth and optical loss in an InP-organic hybrid (IOH) optical modulator, comparing it with a Si-organic hybrid (SOH) optical modulator. By incorporating the RF transmission line model, which enables a more precise modulation bandwidth analysis than the RC constant model, we demonstrate that the IOH modulator can achieve a modulation bandwidth of over 500 GHz with a 2 dB loss, capitalizing on the higher electron mobility of InP. In contrast, the SOH modulator cannot attain a 200 GHz modulation bandwidth with acceptable optical loss. Furthermore, we explore the potential for further enhancing the modulation bandwidth of the IOH modulator by shortening its length, making the IOH modulator a promising candidate for future ultra-high-speed optical modulation.
“…However, electrical packet-switched networks are just one technology. Recent years have given rise to optical circuit switches (OCSes), a newer technology with new possibilities [7], [8], [9]. OCS and other optical technologies, such as those based on wavelength gratings as in Sirius [10], enable dynamic reconfigurable network topologies via optical circuit switching.…”
Section: Introductionmentioning
confidence: 99%
“…Using these optical switches, the topology can change dynamically in relatively short periods and adapt to different traffic patterns, thus facilitating the emergence of reconfigurable datacenter networks (RDCNs) [7], [11]. RDCNs require what was recently termed "Topology engineering" [8]: a new dimension of network design that controls the dynamic topology of the network. Essentially, reconfigurable topologies allow flows to route via shorter paths, increasing throughput by saving network capacity at the cost of a temporal penalty.…”
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.