pFabric: minimal near-optimal datacenter transport

By Mohammad Alizadeh, Shuang Yang, M. Sharif, Sachin Rajsekhar Katti, Nick McKeown, Balaji Prabhakar, Scott Shenker. Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, 2013

945 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Hardware and Architecture, Information Systems

View Paper: https://www.semanticscholar.org/paper/47c141bd9dc9f59090705b6336aed68bfa7ddb47

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 Hardware and Architecture

In Information Systems

Citations in the corpus

Efficient Coflow Scheduling Without Prior Knowledge (2015, methodology): “Non-Blocking Fabric In our analysis, we abstract out the entire datacenter fabric as one non-blocking switch [10, 15, 20, 9, 31, 26] and consider machine uplinks and downlinks as the only sources of contention (Figure 1a).”
Datacenter RPCs can be General and Fast (2018, background): “This is unsurprising: for example, the BDP of the two-tier 10 GbE datacenter used in pFabric is 18 kB [15].”
Swift: Delay is Simple and Effective for Congestion Control in the Datacenter (2020, methodology): “CongestionControlviaPacketScheduling: pFabric[3]achieves near-optimal FCT through the usage of QoS queues.”
Decentralized task-aware scheduling for data center networks (2014, methodology): “Because latency is the primary goal for many data center applications, recent proposals give up on per-ﬂow fairness, and optimize ﬂow-level metrics like meeting ﬂow deadlines and minimizing ﬂow completion time [28, 5].”
CONGA: distributed congestion-aware load balancing for datacenters (2015, methodology): “Similar to prior work [20, 5, 13], we use the flow completion time (FCT) as our main performance metric.”
HULA: Scalable Load Balancing Using Programmable Data Planes (2016, background): “Load balancing also provides the abstraction of a single large output-queued switch for the entire network [1–3], which in turn simpliﬁes bandwidth allocation across tenants [4,5], ﬂows [6], or groups of ﬂows [7].”
Re-architecting datacenter networks and stacks for low latency and high performance (2017, background): “There is a large body of work aiming to tackle various aspects of datacenter transport: most focuses on either achieving low latency [5, 6, 16, 22, 28, 30] or high throughput [2, 12, 31].”
Efficient coflow scheduling with Varys (2014, neutral): “…[15]; but for multiple coﬂows, it minimizes the completion time across all coﬂows and increases the average CCT. Recently proposed shortest-ﬂow-ﬁrst prioritization mechanisms [8,25] (Figure 2b) decrease average FCT, but they increase the average CCT by interleaving ﬂows from different coﬂows.”
SilkRoad: Making Stateful Layer-4 Load Balancing Fast and Cheap Using Switching ASICs (2017, background): “While the community has made e orts to scale out L2/L3 virtual switching to match full bisection bandwidth for intradatacenter tra c (or full gateway capacity for inbound tra c) [17, 30], one missing piece is scaling L4 load balancers to match the full bisection bandwidth of the underlying physical ”
Congestion Control for Large-Scale RDMA Deployments (2015, methodology): “Instead, like prior work [2, 5, 6], we extract salient characteristics of the trace data, such as ﬂow size distribution and generate synthetic trafﬁc to match these characteristics.”
Copa: Practical Delay-Based Congestion Control for the Internet (2018, background): “Many congestion-control algorithms for datacenters exploit the fact that one entity owns and controls the entire network, which makes it easier to incorporate in-network support [1, 3, 29, 36, 17].”
Homa: a receiver-driven low-latency transport protocol using network priorities (2018, methodology): “ew senders to transmit simultaneously. Slightly overcommitting receiver downlinks in this way allows Homa to use network bandwidth efﬁciently: Homa can sustain network loads 2–33% higher than pFabric [4], PIAS, and pHost. Homa limits the overcommitment and integrates it with the priority mechanism t”
Programmable Packet Scheduling at Line Rate (2016, methodology): “For instance, pFabric [12] minimizes flow completion times by coupling a simple switch scheduler (shortest remaining processing time) with a simple end-host protocol (line rate transmission with no congestion control).”