DeTail: reducing the flow completion time tail in datacenter networks

By David Zats, Tathagata Das, Prashanth Mohan, Dhruba Borthakur, Randy Howard Katz. CCRV, 2012

446 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Computer Vision and Pattern Recognition, Information Systems

View Paper: https://www.semanticscholar.org/paper/7d3007ebd72fbbcd00ac4d732feac35a7230f502

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 Computer Vision and Pattern Recognition

In Information Systems

Citations in the corpus

Coflow: a networking abstraction for cluster applications (2012, unknown): “…Systems Cloud Computing General Terms Design Keywords Cluster Networking, Co.ow, Data-Intensive Applications, Datacenter Networks, Cloud Computing 1 Introduction Cluster computing applications serve diverse computing requirements, and they expect a broad spectrum of services from the network.”
Homa: a receiver-driven low-latency transport protocol using network priorities (2018, background): “We focus on tail message latency (99th percentile), as it is the most important metric for datacenter applications [2, 34].”
Low Latency Geo-distributed Data Analytics (2015, background): “3) Optimizing communication patterns: Flow schedulers like D3 [55], PDQ [35], DeTail [62], and D2TCP [51] aim to improve flow completion times or guarantee deadlines.”
pFabric: minimal near-optimal datacenter transport (2013, background): “Better load balancing of course reduces hotspots and thus helps reduce flow completion time, however the techniques and goals are orthogonal and complementary to pFabric.”
Packet Transactions: High-Level Programming for Line-Rate Switches (2015, background): “This rigidity affects many stakeholders adversely: vendors [7, 9, 4] building network switches with merchantsilicon chips [13, 14, 21], network operators deploying switches [85, 80, 58], and researchers developing new switch algorithms [64, 74, 96, 99, 60].”
On the impact of packet spraying in data center networks (2013, methodology): “Thus, future data center designs are likely to ensure low and uniform latencies, using mechanisms such as HULL [11], DCTCP [10], DeTail [14].”
Presto: Edge-based Load Balancing for Fast Datacenter Networks (2015, background): “Unfortunately, popular load balancing schemes based on ﬂow hashing, e.g. , ECMP, cause congestion when hash collisions occur [3, 17, 19, 22, 53, 54, 63] and perform poorly in asymmetric topologies [4, 65].”
HPCC: high precision congestion control (2019, background): “However, it requires non-trivial implementations on switches for Phantom queues to get feedback before link saturation; DeTail [42] needs a new switch architecture for lossless fabric and performs per-packet adaptive load balancing of packet routes in switches; HOMA [33] and NDP [22] are…”
CONGA: distributed congestion-aware load balancing for datacenters (2015, background): “DeTail [52] proposes a per-packet adaptive routing scheme that handles asymmetry using layer-2 back-pressure, but requires end-host modifications to deal with packet reordering.”
Finishing flows quickly with preemptive scheduling (2012, background): “Unlike DeTail which removes the tail, PDQ can save ~30% .ow completion time on average (compared with TCP and RCP), reducing the completion time of almost every .ow (e.g., 85% - 95% of the .ows, Figure 8e).”
RDMA over Commodity Ethernet at Scale (2016, methodology): “RDMA provides a diﬀerent approach.Compared to [41] which still uses TCP, RDMA bypasses the OS kernel, so that the latency introduced by the kernel is eliminated.”
HULA: Scalable Load Balancing Using Programmable Data Planes (2016, background): “DeTail [40] proposes a per-packet adaptive load balancing scheme that adapts to topology asymmetry but requires a complex cross-layer network stack including end-host modifications.”