Deadline-aware datacenter tcp (D2TCP)

Abstract

An important class of datacenter applications, called Online DataIntensive (OLDI) applications, includes Web search, online retail, and advertisement. To achieve good user experience, OLDI applications operate under soft-real-time constraints (e.g., 300 ms latency) which imply deadlines for network communication within the applications. Further, OLDI applications typically employ tree-based algorithms which, in the common case, result in bursts of children-to-parent traffic with tight deadlines. Recent work on datacenter network protocols is either deadline-agnostic (DCTCP) or is deadline-aware (D 3 ) but suffers under bursts due to race conditions. Further, D 3 has the practical drawbacks of requiring changes to the switch hardware and not being able to coexist with legacy TCP.We propose Deadline-Aware Datacenter TCP (D 2 TCP), a novel transport protocol, which handles bursts, is deadline-aware, and is readily deployable. In designing D 2 TCP, we make two contributions: (1) D 2 TCP uses a distributed and reactive approach for bandwidth allocation which fundamentally enables D 2 TCP's properties. (2) D 2 TCP employs a novel congestion avoidance algorithm, which uses ECN feedback and deadlines to modulate the congestion window via a gamma-correction function. Using a small-scale implementation and at-scale simulations, we show that D 2 TCP reduces the fraction of missed deadlines compared to DCTCP and D 3 by 75% and 50%, respectively.

DC-Vegas: A delay-based TCP congestion control algorithm for datacenter applications

“…Most of the experimental setups in this subsection follow previous datacenter TCP studies [14,10,8,15]. …”

“…The deadline missing rates of different algorithms for different fan-in servers are shown in Figure 11. The missed deadline rate is markedly reduced in the DC-Vegas case, which could significantly improve the user experience of OLDI applications [11,15]. Figure 12 plots the aggregated throughput of elephant flows for different algorithms.…”

“…ECN* [9] tries to achieve a performance comparable with DCTCP by only modifying switch ECN marking strategy. Deadline-aware Data Center TCP (D 2 TCP) [15] adds deadline awareness to DCTCP using a gamma correction function. L 2 DCT [16] introduces Least Attained Service (LAS) scheduling into DCTCP to reduce completion times of short flows.…”

“…Incast scenarios widely exists in datacenter applications, such MapReduce [6,7]. To address the soft real-time and incast problems, many improved solutions have been proposed, including fine-grained retransmissions [8], ECN* [9], and ICTCP [10] for mitigating the TCP incast problem, HULL [2], D 3 [11], Detail [12], and PDQ [13] for reducing low transmission latency, as well as DCTCP [14], D 2 TCP [15], L 2 DCT [16], RDT [17] and WDCTCP [18] for both latency reduction and incast avoidance. Although these solutions achieved excellent…”

DAQ: Deadline-Aware Queue scheme for scheduling service flows in data centers

“…The average short flow size is 15 Kbytes and the number of concurrent senders is varied from 5 to 40. The load of short flows is small, up to 0.3% the link capacity [16]. Figure 7a shows the application throughput of short flows.…”