Circuit-switched technologies have long been proposed for handling high-throughput traffic in datacenter networks, but recent developments in nanosecond-scale reconfiguration have created the enticing possibility of handling low-latency traffic as well. The novel Oblivious Reconfigurable Network (ORN) design paradigm promises to deliver on this possibility. Prior work in ORN designs achieved latencies that scale linearly with system size, making them unsuitable for large-scale deployments. Recent theoretical work showed that ORNs can achieve far better latency scaling, proposing theoretical ORN designs that are Pareto optimal in latency and throughput.
In this work, we bridge multiple gaps between theory and practice to develop Shale, the first ORN capable of providing low-latency networking at datacenter scale while still guaranteeing high throughput. By interleaving multiple Pareto optimal schedules in parallel, both latency- and throughput-sensitive flows can achieve optimal performance. To achieve the theoretical low latencies in practice, we design a new congestion control mechanism which is best suited to the characteristics of Shale. In datacenter-scale packet simulations, our design compares favorably with both an in-network congestion mitigation strategy, modern receiver-driven protocols such as NDP, and an idealized analog for sender-driven protocols. We implement an FPGA-based prototype of Shale, achieving orders of magnitude better resource scaling than existing ORN proposals. Finally, we extend our congestion control solution to handle node and link failures.
more »
« less
A Throughput-Centric View of the Performance of Datacenter Topologies
While prior work has explored many proposed datacenter designs, only two designs, Clos-based and expander-based, are generally considered practical because they can scale using commodity switching chips. Prior work has used two diferent metrics, bisection band- width and throughput, for evaluating these topologies at scale. Little is known, theoretically or practically, how these metrics relate to each other. Exploiting characteristics of these topologies, we prove an upper bound on their throughput, then show that this upper bound better estimates worst-case throughput than all previously proposed throughput estimators and scales better than most of them. Using this upper bound, we show that for expander-based topologies, unlike Clos, beyond a certain size of the network, no topology can have full throughput, even if it has full bisection band- width; in fact, even relatively small expander-based topologies fail to achieve full throughput. We conclude by showing that using throughput to evaluate datacenter performance instead of bisection bandwidth can alter conclusions in prior work about datacenter cost, manageability, and reliability.
more »
« less
- Award ID(s):
- 1827977
- NSF-PAR ID:
- 10287681
- Date Published:
- Journal Name:
- ACM SIGCOMM 2021 Conference
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Most recent datacenter topology designs have focused on performance properties such as latency and throughput. In this paper, we explore a new dimension, life cycle management complexity, which attempts to understand the complexity of deploying a topology and expanding it. By analyzing current practice in lifecycle management, we devise complexity metrics for lifecycle management, and show that existing topology classes have low lifecycle management complexity by some measures, but not by others. Motivated by this, we design a new class of topologies, FatClique, that, while being performance-equivalent to existing topologies, is comparable to, or better than them by all our lifecycle management complexity metrics.more » « less
-
Analog integrated circuit (IC) placement is a heavily manual and time-consuming task that has a significant impact on chip quality. Several recent studies apply machine learning (ML) techniques to directly predict the impact of placement on circuit performance or even guide the placement process. However, the significant diversity in analog design topologies can lead to different impacts on performance metrics (e.g., common-mode rejection ratio (CMRR) or offset voltage). Thus, it is unlikely that the same ML model structure will achieve the best performance for all designs and metrics. In addition, customizing ML models for different designs require more tremendous engineering efforts and longer development cycles. In this work, we leverage Neural Architecture Search (NAS) to automatically develop customized neural architectures for different analog circuit designs and metrics. Our proposed NAS methodology supports an unconstrained DAG-based search space containing a wide range of ML operations and topological connections. Our search strategy can efficiently explore this flexible search space and provide every design with the best-customized model to boost the model performance. We make unprejudiced comparisons with the claimed performance of the previous representative work on exactly the same dataset. After fully automated development within only 0.5 days, generated models give 3.61% superior accuracy than the prior art.more » « less
-
The bandwidth and latency requirements of modern datacenter applications have led researchers to propose various topology designs using static, dynamic demand-oblivious (rotor), and/or dynamic demand-aware switches. However, given the diverse nature of datacenter traffic, there is little consensus about how these designs would fare against each other. In this work, we analyze the throughput of existing topology designs under different traffic patterns and study their unique advantages and potential costs in terms of bandwidth and latency ''tax''. To overcome the identified inefficiencies, we propose Cerberus, a unified, two-layer leaf-spine optical datacenter design with three topology types. Cerberus systematically matches different traffic patterns with their most suitable topology type: e.g., latency-sensitive flows are transmitted via a static topology, all-to-all traffic via a rotor topology, and elephant flows via a demand-aware topology. We show analytically and in simulations that Cerberus can improve throughput significantly compared to alternative approaches and operate datacenters at higher loads while being throughput-proportional.more » « less
-
Datacenters need networks that support both low-latency and high-bandwidth packet delivery to meet the stringent requirements of modern applications. We present Opera, a dynamic network that delivers latency-sensitive traffic quickly by relying on multi-hop forwarding in the same way as expander-graph-based approaches, but provides near-optimal bandwidth for bulk flows through direct forwarding over time-varying source-to-destination circuits. Unlike prior approaches, Opera requires no separate electrical network and no active circuit scheduling. The key to Opera's design is the rapid and deterministic reconfiguration of the network, piece-by-piece, such that at any moment in time the network implements an expander graph, yet, integrated across time, the network provides bandwidth-efficient single-hop paths between all racks. We show that Opera supports low-latency traffic with flow completion times comparable to cost-equivalent static topologies, while delivering up to 4x the bandwidth for all-to-all traffic and supporting up to 60% higher load for published datacenter workloads.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
