CONGA: distributed congestion-aware load balancing for datacenters

By Mohammad Alizadeh, Tom Edsall, Sarang Dharmapurikar, Ramanan Vaidyanathan, Kevin Chu, Andy Fingerhut, V. Lam, Francis Matus, Rong Pan, Navindra Yadav and 1 more. Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, 2015

878 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/cb967a0aa79899d1d9559435895827464dfa6761

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

HULA: Scalable Load Balancing Using Programmable Data Planes (2016, background): “CONGA [13] is a recent dataplane load-balancing technique that overcomes ECMP’s limitations by using link utilization information to balance load across paths.”
Re-architecting datacenter networks and stacks for low latency and high performance (2017, background): “At the other end of the spectrum, researchers have been keen to tackle collisions due to per-flow ECMP with centralized scheduling in Hedera [2], via protocol changes in Packet Spraying [12] and Presto [21], or via network subflow-treatment in Conga [3] and switch-redesign of LocalFlow [35].”
Language-Directed Hardware Design for Network Performance Monitoring (2017, methodology): “We evaluate Marple’s usability through two Mininet case studies that use Marple to troubleshoot high tail latencies [26] and measure the distribution of flowlet sizes [27].”
Packet Transactions: High-Level Programming for Line-Rate Switches (2015, background): “An important requirement for data-plane algorithms is the ability to process packets at the switch’s line rate (typically 10–100 Gbit/s on 10–100 ports).”
PINT: Probabilistic In-band Network Telemetry (2020, background): “switchID Dynamic per-flow aggregation Utilization-aware Routing [2, 41, 42] Load balance traffic based on network status.”
Inside the Social Network's (Datacenter) Network (2015, methodology): “…many concurrent ﬂows (§6) Bimodal ACK/MTU packet size, on/off behavior [12]; <5 concurrent large ﬂows [8] SDN controllers [1, 22, 28, 32, 34]; Circuit/hybrid switching [7, 20, 30, 39] Table 1: Each of our major ﬁndings differs from previously published characterizations of datacenter trafﬁc.”
AuTO: scaling deep reinforcement learning for datacenter-scale automatic traffic optimization (2018, methodology): “We adopt the tra c generator [20] used in prior works [2, 7, 9, 15] that produces network tra c ows based on given ow size distribution and tra c load.”
DRILL: Micro Load Balancing for Low-latency Data Center Networks (2017, background): “In CONGA, as another example, each edge switch balances flowlets [19] based on cross-network load information.”
Efficient Coflow Scheduling Without Prior Knowledge (2015, methodology): “Both can improve from in-network support as used in CONGA [9].”
Homa: a receiver-driven low-latency transport protocol using network priorities (2018, background): “…to prior work, we assume that bandwidth in the network core is sufﬁcienttoaccommodatetheofferedload, andthatthenet-work supports efﬁcient load-balancing [10, 16, 1], so that packets are distributed evenly across the available paths (we assume simple randomized per-packet spraying in our design).”
Presto: Edge-based Load Balancing for Fast Datacenter Networks (2015, background): “MPTCP has higher loss because of its bursty nature [4] and its aggression in the face of loss: when a single loss occurs, only one subﬂow reduces its rate.”