Changes in network state are a common source of instability in networks. An update event typically involves multiple flows that compete for network resources at the cost of rescheduling and migrating some existing flows. Previous network updating schemes tackle such flows independently, rather than as the entity of an update event. They only optimize the flow-level metrics for the flows involved in an update event. In this paper, we present an event-level abstraction of network update that groups flows of an update event and schedules them together to minimize the event completion time (ECT). We then study the scheduling problem of multiple update events for achieving high scheduling efficiency and preserving fairness. The designed least migration traffic first (LMTF) method schedules all update events in the FIFO order, but it avoids head-of-line blocking by randomly fine-tuning the queue order of some events. It can considerably reduce the update cost, the average, and tail ECTs of update events. In addition, we design a general parallel-LMTF (P-LMTF) method to guarantee fairness and further improve scheduling efficiency among update events. This improves the LMTF method by opportunistically updating multiple events simultaneously. The comprehensive evaluation results indicate that the average ECT of our approach is up to 10× faster than the flow-level scheduling method for network update events, and its tail ECT is up to 6× faster. Our P-LMTF method incurs a 75% reduction in the average ECT compared with FIFO when the network utilization exceeds 70%, and it achieves a 42% reduction in tail ECT.

In datacenter networks, flows need to complete as quickly as possible because the flow completion time (FCT) directly impacts user experience, and thus revenue. Link failures can have a significant impact on short latency-sensitive flows because they increase their FCTs by several fold. Existing link failure management techniques cannot keep the FCTs low under link failures because they cannot completely eliminate packet loss during such failures. We observe that to completely mask the effect of packet loss and the resulting long recovery latency, the network has to be responsible for packet loss recovery instead of relying on end-to-end recovery. To this end, we propose Shared Queue Ring (SQR), an on-switch mechanism that completely eliminates packet loss during link failures by diverting the affected flows seamlessly to alternative paths. We implemented SQR on a Barefoot Tofino switch using the P4 programming language. Our evaluation on a hardware testbed shows that SQR can completely mask link failures and reduce tail FCT by up to 4 orders of magnitude for latency-sensitive workloads.

The cloud datacenter network is consistently undergoing changing, due to a variety of topology and traffic updates, such as the VM migrations. Given an update event, prior methods focus on finding a sequence of lossless transitions from an initial network state to an end network state. They, however, suffer frequent and global search of the feasible end network states. This incurs non-trivial computation overhead and decision-making delay, especially in large-scale networks. Moreover, in each round of transition, prior methods usually cause the cascaded migrations of existing flows; hence, significantly disrupt production services in IaaS data centers. To tackle such severe issues, we present a simple update mechanism to minimize the amount of flow migrations during the congestion-free network update. The basic idea is to replace performing the sequence of global transitions of network states with local reschedule of involved flows, caused by an update event. We first model all involved flows due to an update event as a set of new flows, and then propose a heuristic method Lupdate. It motivates to locally schedule each new flow into the shortest path, at the cost of causing the extra migration of at most one existing flow if needed. To minimize the amount of migrated traffic, the migrated flow should be as small as possible. To further improve the success rate, we propose an enhanced method Lupdate-S. It shares the similar design of Lupdate, but permits to migrate multiple necessary flows on the shortest path allocated to each new flow. We conduct large-scale trace-driven evaluations under widely used Fat-Tree and ER data centers. The experimental results indicate that our methods can realize congestion-free network with as less amount of traffic migration as possible even when the link utilization of a majority of links is very high. The amount of traffic migration caused by our Ludpate method is 1.2 times and 1.12 times of the optimal result in the Fat-Tree and ER random networks, respectively.

Changes of network state are a common source of instability in networks. An update event typically involves multiple flows that compete for network resources at the cost of rescheduling and migrating some existing flows. Previous network updating schemes tackle such flows independently, rather than as the entity of an update event. They only optimize the flow-level metrics for the flows involved in an update event. In this paper, we present an event-level abstraction of network update which groups flows of an update event and schedules them together to minimize the event completion time (ECT). We then study the scheduling problem of multiple update events for achieving high scheduling efficiency and preserving fairness. The designed least migration traffic first (LMTF) method schedules all update events in the FIFO order, but avoids head-of-line blocking by randomly fine-tuning the queue order of some events. It can considerably reduce the update cost, the average, and tail ECTs of all update events. In addition, we design a general parallel-LMTF (P-LMTF) method to guarantee fairness and further improve scheduling efficiency among update events. It improves the LMTF method by opportunistically updating multiple events simultaneously. The comprehensive evaluation results indicate that the average ECT of our approach is up to 10× faster than the flow-level scheduling method for network update events, and its tail ECT is up to 6x faster. Our P-LMTF method incurs 75% reduction in the average ECT compared with FIFO when the network utilization exceeds 70%, and it achieves a 42% reduction in tail ECT.

Software-Define Datacenter networks (SDDCs) are constantly changing due to various network update events. A set of involved flows of each update event should be migrated to the feasible paths without congestion. To tackle a single update event, prior methods find and execute a migration sequence so as to transform the initial traffic distribution to the final traffic distribution. However, when handling a queue of multiple update events, the head-of-line blocking problem always appears and considerably lower the efficiency of network update. To tackle this problem, we employ the idea of delay updating which schedules other queued events instead of the blocked head-event, aiming to break the blocking state and provide updating opportunities for other events. We formulate this problem as an optimization problem and propose partial delay updating (PDU) strategy. PDU tackles the queued update events based on their arrival order to preserve the fairness. Moreover, it prefers to just delay those blocked flows, lacking enough bandwidth resources, instead of all flows in the head-event. The evaluation results show that our delay updating strategy achieves 60%–80% reduction in average completion time of update events, compared to FIFO scheduling strategy, in four types of flow size.

Recent results have made a promising case for offering oversubscribed wired data center networks (DCN) with extreme costs. Inter-rack wireless networks are drawing intensive attention to augment such wired DCNs with a few wireless links. Inspired by the promise of easy deployment and plug-and-play, we present VLCcube, a novel inter-rack wireless solution that extends the design of wireless DCN into three further dimensions: (1) all inter-rack links are wireless; (2) there is no imposition of any infrastructure-level alteration on wired production data centers; and (3) it should be plug-and-play, without any need of additional mechanical or electronic control operations. This vision, if realized, will lead to increased flexibility, reduced reconstructing cost, simplified configuration and usage, and outstanding compatibility with existing wired DCNs. Previous proposals, however, are opposed to the last two design rationales. To achieve this vision, the proposed VLCcube augments Fat-Tree, a representative DCN in production data centers, by organizing all racks into a wireless Torus structure via the emerging visible light links. We further present the topology design, hybrid routing, and flow scheduling schemes for VLCcube. Extensive evaluations indicate that VLCcube outperforms Fat-Tree significantly under the existing ECMP flow scheduling scheme, irrespective of the undergoing traffic pattern. Moreover, the performance of VLCcube can be significantly promoted by our congestion-aware flow scheduling scheme. More precisely, compared to ECMP, our flow scheduling scheme makes VLCcube achieve$\times 1.50$throughput under batched flows,$\mathrm{\times}2.21$and$\times 2.59$throughput under two different kinds of online flows.

Summary Large‐scale online services and distributed execution engines (i.e., MapReduce and Dryad) generate large volumes of traffic in data center networks. As a consequence, significant congestion can occur in the data center network. To the best of our knowledge, most existing approaches either focus on local congestion‐aware mechanisms, which have only a poor ability to handle asymmetry or use explicit congestion notification packets, which are difficult to implement directly in switch hardware. These methods are insufficient to solve the congestion problem. In this paper, we focus on a congestion‐free routing strategy, resorting to the global view of the data center network in a software‐defined networking controller. Specifically, a timeslot allocation was first conducted for the coming packets, and then the corresponding routing paths were computed for each packet. In view of the efficiency, the timeslot allocation algorithm follows a heuristic pattern, and the path selection is modeled as a bin‐packing problem. Simulation results showed that the congestion‐free routing strategy proposed here performs well in throughput, queuing, and end‐to‐end round‐trip time. Copyright © 2015 John Wiley & Sons, Ltd.