- Author: Habtegebreil Haile
- Index: ITC(International Teletraffic Congress) 2018
- Reading date: 09/10/2019
- Keywords: Evaluation, CUBIC, BBR
new alternative algorithm requires a thorough evaluation of the effect of the proposed alternative on established transport protocols like TCP CUBIC. Evaluations that consider the heterogeneity of Internet traffic sizes. Most evaluations before of BBR focus on steady-state performance of large flows.The steady-state operation of BBR can have negative impact on the performance of bursty flows using loss-based CCAs.
- Author: Unknown
- Index: introduce
- Reading date: 09/10/2019
- Keywords: Congestion Avoidance / Fast Retransmit
Congestion Detection/ loss-based/ buffer bloat / MTU vs MSS. / send window / initial window, IW .
Basic introduction to these concepts. e.g. (Retransmission Timeout), (Fast Retransmission).
MSS (maximum segment size) 536 bytes. MTU(maximum transmission unit)
Ethernet MTU=1500 MSS=MTU-20 octet(TCP)-20 octet (IP) this two unit for congestion control.
SMSS;RMSS;Flight Size; Intiral window = 2SMSS-4SMSS ; 2013 10SMSS
- Author: Mowei Wang and Junchen Jiang
- Index: IEEE Network
- Reading date: 12/07/2019
- Keywords: Machine Learning and Network
Networking itself can benefit from this promising technologyββMachine Learning. It can not only help solve the intractable old network questions but also simulate new network applications. Finally, this paper shed light on the new opportunities in networking design and community building of this new inter-discipline.
- Author: Xiaohui Nie
- Index: JSAC (Journal paper)
- Reading date: 22/04/2019
- Keywords: Congestion Control Machine Learning
This paper use NS3 and DeepReinforcement Learning to build a smart congestion control algorithm. It is called TCP-Deep ReInforcementlearNing-based Congestion control (Drinc) which learns from past experience in the form of a set of measured features to decide how to adjust the congestion window size. TCP-Drinc has a good result.
- Author: Juncheng Gu
- Index: NSDI 2019
- Reading date : 09/03/2019
- Keywords: Distributed System; Machine Learning
Tiresias, a GPU cluster manager tailored for distributed DL training jobs, which efficiently schedules and places DL jobs to reduce their job completion times (JCTs).Given that a DL jobβs execution time is often unpredictable,
- we propose two scheduling algorithms β Discretized Two Dimensional Gittins index relies on partial information and Discretized Two-Dimensional LAS is information-agnostic that aim to minimize the average JCT.
- Additionally, we describe when the consolidated placement constraint can be relaxed, and present a placement algorithm to leverage these observations without any user input.
- Author: Yang Qiang
- Index: arXiv:1812.03337
- Reading date : 09/03/2019
- Keywords: Federal Learning
ML relies on large amount of data for training. However, most data are scattered and not shared by company A & B. In this paper, we introduce a federated transfer learning(FTL) let (target-domain) party to use more (source-domain) party labels.
It use Encryption Algorithm to ensure this.
- Author: Wei Bai
- Index: NSDI 2015
- Reading date : 28/02/2019
- Keywords : Flow scheduling, Data center networks
PIAS is a flow scheduling algorithm whose target is minimize FCT without prior knowledge. And we have implemented a PIAS prototype and evaluated PIAS through both testbed experiments and NS-2 simulations.
so the three key design goals :
-
information-agnostic (unlike oher PDQ pFabric and PASE)
-
FCT minimization
-
Readily-deployable (DCTCP HULL L2DCT) reduce FCT without relying on flow size information by ECN pacing (end-host based rate control ineffective)
-
how to determine the demoted threshold for each queue of MLFQ
-
as flow size distribution varies across time and space. how to keep PIAS's performance in such a dynamic environment.
-
how to ensure PIAS's compatibility with legacy TCP/IP stacks in production DCNs?
- Author: Nandita Dukkipati
- Index: SIGCOMM Computer Communication Rebiew
- Reading date : 26/02/2019
- Keywords : Congestion Control
Users typically want their flows to complete as quickly as possible. This makes Flow Completion Time(FCT) an important-arguably the most important - performance metric for the user. Yet research on congestion control focuses almost entirely on maximizing link throughput, utilization and fairness, which matter more to the operator than the user.
In this paper we show that with typical Internet flow sizes, existing (TCP Reno) and newly proposed (XCP) congestion control algorithms make flow last much longer than necessary. Often by one or two orders of magnitude.
State-of-the-Art Deep Learning: Evolving Machine Intelligence Toward Tomorrow's Intelligent Network Traffic Control Systems β
- Author: Zubair MD. Fadlullah(Tohoku University)
- Index: IEEE Communications Surveys
- Reading date : 25/02/2019
- Keywords : Machine Learning, Artificial Neural Network
At present, the network traffic control system is mainly composed of an Internet core and a wired/wireless heterogeneous backbone network. Recently, due to the rapid development of communication technologies, these packet switching systems are experiencing explosive network traffic growth. The existing network strategy is not complex enough to cope with the ever-changing network traffic with huge traffic growth.
In this paper, we have solved this and pointed out the necessity of investigating the decentralized work of deep learning and various network traffic control applications. In this regard, we outline the most advanced deep learning architectures and algorithms associated with network traffic control systems. In addition, we also discussed the driving force of deep learning - the network system.
In addition, we discuss in detail a new use case, intelligent routing based on deep learning. We demonstrate the effectiveness of deep learning-based routing methods compared to traditional routing strategies.
- Author: Kefan Xiao Shiwen Mao
- Index: IEEE Access
- Reading date : 25/02/2019
- Keywords : congestion control, deep convolutional neural network(DCNN)
This is one of the three important papers i refer to for reproducing.
In this paper, we develop a model free,smart congestion control algorithm based on deep reinforcement learning, which has a high potential in dealing with the complex and dynamic network environment. We present TCP-Deep ReInforcement learNing-based Congestion control (Drinc) which learns from past experience in the form of a set of measured features to decide how to adjust the congestion window size. We present the TCP-Drinc design and validate its performance with extensive ns-3 simulations and comparison with benchmark schemes.
State-of-the-Art Deep Learning: Evolving Machine Intelligence Toward Tomorrowβs Intelligent Network Traffic Control Systems β
-
Author: Zubair Md.Fadlullah
-
Index: IEEE 2017
-
Reading date : 25/02/2019
-
Keywords : Machine learning, Artificial Neural Network, Network Traffic Control
- Author: Wei Li (NorthEastern University)
- Index: DCOSS 2016
- Reading date : 24/02/2019
- Keywords : TCP, IoT, Q-learning, Kanerva coding
I read again the author Wei Li's paper about using q-learning (rl) to tcp congestion control. This is sequence two and In coincidence, this idea is very much similar to mine.
Advances in wireless technology have resulted in pervasive deployment of devices of a high variability in form factors, memory and computational ability. The need for maintaining continuous connections that deliver data with high reliability necessitate re-thinking of conventional design of the transport layer protocol.
This paper investigates the use of Q-learning in TCP cwnd adaptation during gthe congestion avoidance state, wherein the classical alternation of the window is replaced, thereby allowing the protocol to immediately respond to previously seen network conditions. We reduce the memory requirements of a learning-based protocol while maintaining the same throughput and delay.
HyperLoop: Group-Based NIC-Offloading to Accelerate Replicated Transactions in Multi-Tenant Storage Systems β
- Author: Daehyeok Kim(CMU Computer Science) Yibo Zhu (Bytedance)
- Index: SIGCOMM 2018
- Reading date : 23/02/2019
- Keywords : Distributed Storage System, Replicated transactions, RDMA NIC-offloading
Storage systems in data centers are an important component of large-scale online services. They typically perform replicated transactional operations for high data availability and integrity.
Today, however, such operations suffer from hightail latency even with recent kernel bypass and storage optimizations(RDMA),and thus affect the predictability of end-to-end performance of these services.We observe that the root cause of the problem is the involvement of the CPU, a precious commodity in multi-tenant settings, in the critical path of replicated transactions.
In this paper, we present HyperLoop, a new framework that removes CPU from the critical path of replicated transactions in storage systems by offloading them to commodity RDMA NICs. To achieve this, We develop new and general NIC offloading primitives that can perform memory operations on all nodes in a replication group while guaranteeing ACID properties without CPU involvement.
- Author: Yizhou Sun(Ph.D. at Purdue ECE)
- Index: OSDI 2018 best paper
- Reading date : 23/02/2019
- Keywords : OS, Kernel
The monolithic server model where a server is the unit of deployment, operation, and failure is meeting its limits in the face of several recent hardware and application trends. To improve resource utilization, elasticity, heterogeneity, and failure handling in datacenters, we believe that datacenters should break monolithic servers into disaggregated, network-attached hardware components. Despite the promising benefits of hardware resource disaggregation, no existing OSes or software systems can properly manage it.
We propose a new OS model called the splitkernel to manage disaggregated systems. Splitkernel disseminates traditional OS functionalities into loosely-coupled monitors, each of which runs on and manages a hardware component. A splitkernel also performs resource allocation and failure handling of a distributed set of hardware components. Using the splitkernel model, we built LegoOS, a new OS designed for hardware resource disaggregation.
- Split OS functions into monitors
- Run each monitor at hardware device
- network messaging across non-coherent components
- global resource managgement and failure handling
- Author: STEFANO PETRANGELI (Ghent University-iMinds)
- Index: ACM Trans October 2015
- Reading date : 22/02/2019
- Keywords : Experimental evaluation, fairness, HTTP Adaptive Streaming
HTTP Adaptive Streaming (HAS) is quickly becoming the de facto standard for video streaming services. In HAS, each video is temporally segmented and stored in different quality levels. Rate adaptation heuristics, deployed at the video player, allow the most appropriate level to be dynamically requested, based on the current network conditions. It has been shown that todayβs heuristics underperform when multiple clients consume video at the same time, due to fairness issues among clients.
- Author: Wei Li (NorthEastern University)
- Index: IEEE Article May 2018
- Reading date : 22/02/2019
- Keywords : Congestion Control, Q-learning
Efforts on optimizing the performance of TCP by modifying the core congestion control method depending on specific network architectures or apps do not generalize well under a wide range of network scenarios. Tranditional method where the performance is linked to a pre-decided mapping between the observed state of the network to the corresponding actions.
- We address this problem by integrating a reinforcement-based Q-learning framework with TCP design in our approach called QTCP. QTCP enables senders to gradually learn the optimal congestion control policy in an on-line manner. QTCP does not need hard-coded rules, and can therefore generalize to a variety of different networking scenarios.
- Moreover, we develop a generalized Kanerva coding function approximation algorithm, which reduces the computation complexity of value functions and the searchable size of the state space.
- Author: Nathan Jay(UIUC University of Illinois at Urbana-Champaign)
- Index: NIPS 2018 Poster
- Reading date : 21/02/2019
- Keywords : Congestion Control, Reinforcement Learning
We present and investigate a novel and timely application domain for deep reinforcement learning (RL): Internet congestion control.Congestion control is fundamental to computer networking research and practice in light of the advent of Internet services such as live video, augmented and virtual reality, Internet-of-Things, and more. Performance-oriented Congestion Control (PCC) framework to formulate congestion control protocol design as an RL task.We present and discuss the challenges that must be overcome so as to realize our long-term vision for Internet congestion control.
- Author: Akshay Aggrawal
- Index: CS211 Final Project
- Reading date: 18/02/2019
- keywords: CC , RL , NS2
In this paper, i present Xaiver, a congestion control policy informed by reinforcement learning and a first step towards adaptible control and analyze its perfromance on two simulated network topologies. Experimental results hint at the utility of using reinforcement-learning for congestion control while also revealing difficulites entailed by doing so.
PCC-244 X 3 : PCC Shallow Queues and Fairness / PCC fairness and flow completion time / Re-architectingg CC for high performance
- Author: Stanford Students
- Index: π£
- Reading date: 17/02/2019
- Keywords: PCC
These three papers are PCC related. The reason why i read these three paper is for their experiment settings.
Following are what they got for the key insights:
-
1.PCC performance degrades in virtual environments under certain conditions. However even in the environments multiple PCC flows show significantly greater stable convergence and faireness when compared with typical TCP flows.
-
2.What the authors found was that PCC outperforms specially-engineered TCP algorithms in multiple scenarios on a variety of metrics which proves the merit of PCC and the lacking of TCP in specific scenarios. PCC has faster and better convergence to flow-fairness, as seen in figures 6, 12, and 13, and it utilizes buffers better, as seen in figures 4 and 7. What the latter result implies is that it is possible to build routers with very limited memory. Furthermore, in figure 15 we can see that in the worst-case these results are not achieved in the expense of short flow completion time as, even though PCC is slower than CUBIC, it is only 20% slower in the worst case.
-
3.PCC with the default utility function indeed does provide far better throughput than TCP in suboptimal network conditions. However, it also is very unfriendly to existing TCP flows and has significant problems with operating in a virtualized setting that shares tenancy. We discuss the tradeoffs and offer our critique below.
- Author: Google
- Reading date : 16/02/2019
- Keywords : protocol
QUIC(Quick UDP Internet Connections) is a new transport protocol for the internet, developed by Google. QUIC solves a number of transport-layer and application-layer problems experienced by modern web applications, while requiring little or no change from application writers.
QUIC is very similar to TCP+TLS+HTTP2, but implemented on top of UDP. Having QUIC as a self-contained protocol allows innovations which aren't possible with existing protocols as they are hampered by legacy clients and middleboxes.
Key advantages of QUIC over TCP+TLS+HTTP2 include:
- Connection establishment latency
- Improved congestion control
- Multiplexing without head-of-line blocking
- Forward error correction
- Connection migration
- Author: Francis Y. Yanβ , (Stanford University)
- Index: Best Paper at USENIX ATC 2018
- Reading date : 15/02/2019
- Keywords : Congestion Control, Machine Learning
Indigo is a machine-learned congestion-control scheme whose design we extract from data gathered by Pantheon.Pantheon's calibrated emulators provide an alternative: they are reproducible, can be instantiated many times in parallel. Our high-level strategy is to train Indigo using emulators, then evaluate it in the real world using Pantheon.
Pantheon facilitates realistic offline training and testing by providing a communal benchmark, evolving dataset, and calibrated emulators to allow approximately realistic offline training and testing.
- Author: Junchen Jiang (CMU,Computer Science at The University of Chicago AP)
- Index: Thesis
- Reading date : 12/02/2019
- Keywords : Video Streaming, Future research
'Eyeball economy' is dominated by applications such as video streaming and VoIP, maintaining high user-perceived QoE(Quality of Experience) has become crucial to ensure high user engagement. recent research shows even one short buffering interruption leads to 0.39 less time. They either require costly re-architecting of the network core or use suboptimal endpoint-based protocols to react to the dynamic Internet performance based on limited knowledgge of the network.
Author demonstrate that data-driven techniques can improve Internet QoE by utilizing a centralized real-time view of performance across millions of endpoints(clients). Work focus on two fundamental challenges that are unique to applying data-driven approaches in networking:
- the need for expressive models to capture complex factors affecting QoE
- the need for scalable platforms to make real-time decisions with fresh data from geo-distributed clients.
Author address these challenges by integrating several domain-specific insights in networked applications with ML algorithms and systems. And achieve better QoE than using off-the-shelf machine learning solutions.
- Author: Pascal_Poupart_(Waterloo), Li Chen(HKUST),Kai Chen (HKUST)
- Index: ICNP 2016
- Reading date : 15/02/2019
- Keywords : Networks, Online flow size prediction, Routing
ICNP (The 26th IEEE International Conference on Network Protocols)/ We describe an emergint application o data mining in the context of computer networks. This application concerns the problem of predicting the size of a flow and detecting elephant flows(very large flows). Flow size is a very important statistic that can be used to improve routing, load balancing and scheduling in computer networks.
Flow size prediction is particularly challenging since flow patterns continuously change and predictions must be done in real time (milliseconds) to avoid delays.
We evaluate the accuracy of three online predictors based on neural networks, Gaussian process regression and online Bayesian Moment Matching on three datasets of real traffic.
We also demonstrate how to use such online predictors to improve routing (i.e., reduced flow completion time) in a network simulation.
- Author: JunZhang (Deakin University)
- Index: IEEE August 2015
- Reading date : 11/02/2019
- Keywords : Traffic Classification, Semi-supervised Learning, Zero-day applications
As a fundamental tool for network management and security, traffic classification has attracted increasing attention in recent years.
A significant challenge to the robustness of classification performance comes from zero-day applications previously unknown in traffic classification systems.
In this paper, we propose a new scheme of Robust statistical Traffic Classification (RTC) by combining supervised and unsupervised machine learning techniques to meet this challenge.
-
The proposed RTC scheme has the capability of identifying the traffic of zero-day applications as well as accurately discriminating predefined application classes.
-
In addition, we develop a new method for automating the RTC scheme parameters optimization process. The empirical study on real-world traffic data confirms the effectiveness of the proposed scheme.
When zero-day applications are present, the classification performance of the new scheme is significantly better than four state-of-the-art methods:
- random forest
- correlation-based classification
- semi-supervised clustering,
- one-class SVM.
- Author: Junchen Jiang (CMU,,Computer Science at The University of Chicago AP)
- Index: NSDI 2016
- Reading date : 11/02/2019
- Keywords : Video Streaming, Bitrate Adaption
Network operators measure Internet routes to troubleshoot problems, and researchers measure routes to characterize the Internet. However they still rely on decades-old tools like traceroute, BGP route collectors and looking Glasses.
This paper presents Sibyl, a system that takes rich queries that researchers and operators express as regular expressions, then issues and returns traceroutes that match even if it has never measured a matching path in the past.
There are 3 steps:
-
to maximize its coverage of Internet routing, Sibyl integrates together diverse sets of traceroute vantage points that provide complementary views.
-
Users may not know which measurements will traverse paths of interest.
-
Sibyl optimizes across concurrent queries to decide which measurements to issue given resource constraints.
MLinSDN: Machine Learning in Software Defined Networks: Data Collection and Traffic Classification β
- Author: Pedro Amaral
- Index: LoCoSDN project
- Reading date : 21/02/2019
- Keywords : Software defined Networks; Machine Learning; Data Analysis; Traffic Classification
AbstractβSoftware Defined Networks (SDNs) provides a separation between the control plane and the forwarding plane of networks. The software implementation of the control plane and the built in data collection mechanisms of the OpenFlow protocol promise to be excellent tools to implement Machine Learning (ML) network control applications. In this work we describe a simple architecture deployed in an enterprise network that gathers traffic data using the OpenFlow protocol.We present the data-sets that can be obtained and show how several ML techniques can be applied to it for traffic classification. The results indicate that high accuracy classification can be obtained with the data-sets using supervised learning.
- Author: hongzimao (MIT Computer Science and Artificial Intelligence Laboratory)
- Index: HotNets 2016
- Reading date : 21/02/2019
- Keywords : Resource Mangement, Systems and Networks
Resource management problems in systems and networking often manifest as difficult online decision making tasks where appropriate solutions depend on understanding the workload and environment. Inspired by recent advances in deep reinforcement learning for AI problems, we consider building systems that learn to manage resources directly from experience. We present DeepRM, an example solution that translates the problem of packing tasks with multiple resource demands into a learning problem. Our initial results show that DeepRM performs comparably to state-of-the-art heuristics, adapts to different conditions, converges quickly, and learns strategies that are sensible in hindsight.
Pytheas: Enabling Data-Driven Quality of Experience Optimization Using Group-Based Exploration-Exploitation β
- Author: Junchen Jiang (CMU,,Computer Science at The University of Chicago AP)
- Index: NSDI 2016
- Reading date : 11/02/2019
- Keywords : Video Streaming, Bitrate Adaption
Continuing with the Sequence 3.
Many ways of optimizing Quality of Experience (QoE) are using data-driven mechanisms. they use the recent sessions 's QoE result to do the optimal decisions. However, they are not accurate and incomplete due to: (1) suffer from many known biases (2) cannot respond to sudden changes.
the author think that the Drawing a parallel from machine learning, data-driven QoE optimization should instead be cast as a real-time exploration and exploitation (E2) rather than as a prediction problem. Adopting E2 in network applications, however, introduces key architectural challenge and algorithm challenge.
We present Pytheas, a framework which addresses these challenges using a group-based E2 mechanism. The insight is that application sessions sharing the same features can be grouped so that we can run E2 algorithms at a per-group granularity.Using an end-to-end implementation and a proof-of-concept deployment in CloudLab, We show that Pytheas improves video QoE
- Author: Junchen Jiang (CMU,,Computer Science at The University of Chicago AP)
- Index: NSDI 2016
- Reading date : 11/02/2019
- Keywords : Video Streaming, Bitrate Adaption
Continuing with the Sequence 2.
Many prior efforts have suggested that internet video QoE could be dramatically improved by using data-drive prediction of video quality for different choices. (e.g, CDN or bitrate) to make optimal decisions. model to expressive enouggh to capture these complex relationships and capble of updating quality predictions in near real-time. We design and implement CFA(Critical Feature Analytics). This is driven by domain-specific insights hat video quality is typically determined by a small subset of critical features whose criticality persists over several tens of minutes.
learn critical features for different sesions on coarse-grained timescales.
Using a combination of a real-world pilot deplyment and trace-driven analysis. we demonstrate that CFA leads to significant improvements in video quality. 32% less buffering time and 12% higher bitrate than a random decision maker.
- Author: Yi Sun ( CMU Visitor δΈη§ι’(UCAS)AP )
- Index: Sigcomm 2016
- Reading date : 11/02/2019
- Keywords : Internet Video, TCP, Throughput Prediction, Bitrate Adaption, Dynamic Adaptive Streaming over HTTP(DASH)
Continuing with the Sequence 1.
Bitrate adaptation is critical to ensure good quality-of-experience(QoE) for Internet Video.οΌ1) initial bitrate selection lower startup delay and offer high initial resolution (2) mid-stream bitrate adaption for high QoE. Prior efforts did not systematically quantify real-world throughput predictability or develop good prediction algorithms. To bridge this gap, the paper do three contributions:
- Sessions sharing similar key features present similar initial throughput values and dynamic patterns
- There is a natural "stateful" behavior in throughput variability within a given session
with these insights, we build CS2P, a throughput prediction system which use a data-driven approach to learn
- cluster of similar sessions.
- an initial throughput predictor.
- a Hidden-Markov-Model based midstream predictor modeling the stateful evolution of throughput.
we develop a prototype system and show using trace-driven simulation and real-world experiments that:(1) CS2P outperforms existing prediction approaches (2) CS2P achieve improvement on overall QoE and high average bitrate over state-of-art model predictive conrol(MPC).
DBA: Routing or Computing? The Paradigm Shift Towards Intelligent Computer Network Packet Transmission Based on Deep Learning β
- Author: Bomin Mao (Tohoku University ζ₯ζ¬ζ±εε€§εΈ )
- Index:
Journal| IEEE Transactions on Computers 2017 - Reading date : 10/02/2019
- Keywords : Software defined routers, network traffic control, deep learning, backbone networks, routing.
- Software Defined Routers (programmable routers) have emerged as a viable solution.
- Multi-core platforms significantly promote SDRs' parallel computing capacities, enabling them to adopt artificial intelligent techniques. i.e., deep learning, to manage routing paths.
In this paper, we explore the deep learning based route estimation for high-throughput packet processing under the GPU-accelerated SDRs.We envision a supervised deep learning system to construct the routing tables. We show how the proposed method can be integrated with programmable routers using both (CPUs) and (GPUs).we proposed DBA(deep belief architecture) to predict the next nodes. In particular, the simulation results demonstrate that our proposal outperforms the benchmark method in terms of delay, throughput, and signaling overhead.
- Author: Mo Dong and Tong Meng (UIUC)
- Index: NSDI 2018
- Reading date : 10/02/2019
- Keywords : congestion control, reinforcement learning
Continuing with the Sequence 2.
Author leverage ideas from the rich literature on online (convex) optimization in machine learning to design Vivace, a novel rate-control protocol, designed within the recently proposed PCC framework. PCC Vivace outperforms the previous realization of the PCC framework, and BBR in terms of performance (throughput, latency, loss), convergence speed, alleviating bufferbloat, reactivity to changing network conditions, and friendliness towards legacy TCP in a range of scenarios. Vivace requires only sender-side changes and is thus readily deployable.
- Author: Mo Dong (UIUC)
- Index: NSDI 2015
- Reading date : 10/02/2019
- Keywords : congestion control, reinforcement learning
Continuing with the Sequence 1.
Add pcc website
TCP has little hope of achieving high performance due to fundamental architectural deficiency. Propose PCC(Peformance-oriented Congestion Control) PCC in which each sender continuously observes the connection between its actions and empricially experienced performance, enabling it to consistently adopt actions that result in high performance. We prove that PCC convereges to a stable and fair equilibrium.
- Author: Keith Winstein, Hari Balakrishnan (MIT Now is already Stanford's AP teaching CS244)
- Index: Sigcomm 2013
- Reading date : 10/02/2019
- Keywords : congestion control, computer-generated algorithms.
Although i already heard about Remy on the group meeting, but there is no reason to skip it. Since the algorithm used inside the paper is so RL-like.
This paper describe a new apparoach to end-to-end congestion control on a multi-user network. We developed a program called Remy that generates contestion-conrol algorithms to run at the endpoints. Remy produces A distributed algorithm to achieve high throughput and low queueing delay.
In simulations with ns-2, Remy-generated algorithms outperformed human-designed end-to-end techniques, including TCP Cubic, Compound and Vegas. Remy performs better than algorithms in parameters known as priori(datacenter);prior knowledgeg is less precise(cellular network); estimate the sensitivity of the resulting performance to the specificity of the prior knowledge.
- Author: Hongzi Mao (MIT Computer Science and Artificial Intelligence Laboratory)
- Index: Sigcomm 2017
- Reading date : 09/02/2019
- Keywords : bitrate adaptation, video streaming, reinforcement learning
Client-side video players employ ABR algorithms to optimize user QoE. However, state-of-the-art ABR algorithms suffer from limitation: use fixed control rules based on simplified or inaccurate models of the deployment environment. We propose Pensieve, a system that generates ABR alggorithms using reinforcement learning (RL). Pensieve trains a neural network model hat selects bitrates for future video chunks based on observations collected by client video players. No matter that it outperforms the state-of-the-art scheme.
- Author: Haoyu Wang (UVA University of Virginia)
- Index: Infocom 2018
- Reading date : 08/02/2019
- Keywords : Congestion Control, Web application
I read this paper because it is Infocom 2018 conference paper and it is about congestion control. But i have to say this paper's abstraction is not so good for reading.
With rapid development of web applications in datacenter, network latency becomes more important to user experience.The network latency will be greatly increased by incast congestion, in which a huge number of requests arrive at the front-end server simultaneously. This is an incast problem. Previous incast problem solutions are usually not efficient. In this paper, PICC (Proactive Incast Control System) is proposed. It limits the number of data servers concurrently connected to the front-end server to avoid the incast contestion through data placement.PICC incorporate a queuing delay reduction algorithm that assigns higher transmission priority to data objects with smaller size and longer queuing times.
- Author: Akshay Narayan (MIT Computer Science and Artificial Intelligence Laboratory)
- Index: Sigcomm 2018
- Reading date : 08/02/2019
- Keywords : Congestion Control, Operating System
This paper is very special idea contained. What it does is a unite thought. It describes the implementation and evaluation of a system to implement complex congestion control functions by placing them in a separate agent outside the datapath (which means Linux Kernel TCP, UDP-based QUIC, kernel-bypass transports like mTCP-on-DPDK) these datapath is for the information about packet rtt and ECN, losses. now we call this off-datapath system Congestion Control Plane(CCP). "write once, run anywhere" makes it supereasy.
- Author: Ankit Singla (University of Illinois at UrbanaβChampaign)
- Index: HotCloud 2011
- Reading date : 31/01/2019
- Keywords : network topology
This paper is for the DataCenter Networking structure, because the datacenter needs high bandwidth requirement. So itself naturally to incremental expansion. Suprisingly, the Jellyfish is more cost-efficient than a fat-tree. 25% more servers at full capacity using the same equipment at the scale of a few thousand nodes, and this advantage improve with scale. I learn what is HotCloud and the first sample reproducing.
- Author: Francis Y. Yanβ , (Stanford University)
- Index: Best Paper at USENIX ATC 2018
- Reading date : 27/01/2019
- Keywords : Congestion Control
The author present the Pantheon, a system taht addresses this by serving as a community " training ground " for research on Internet Protocols and congestion control. There is a common set of benchmark algorithms and a shared evaluation platform, and a public archive of results. Pantheon can also be used to develop new CC schemes , two of which already published on NSDI 2018. I think this is mysterious and interesting that's why i choose this paper to read.
- Author: Yibo Zhu (Microsoft;ByteDance)
- Index: CoNEXT 2016
- Reading date : 20/01/2019
- Keywords : Datacenter transport, RDMA, ECN, delay-based, congestion control
This is also for [RDMA Section] paper 2. Aiming at reviewing the classical paper in relative to RDMA. Based on the paper name, we already know that this paper is about the compare of DCQCN and TIMELY.
According to the YiboZhu, which is the author of the DCQCN: DC networks especially drop-free RoCEv2 networks require efficient cc protocols. DCQCN(ECN-based) and TIMELY(delay-based) are two recent proposals for this purpose. In this paper, we analyze DCQCN and TIMELY using fluid models and simulations, for stability, convergence, fairness and flow completion time. We uncover several surprising behaviors of these protocols.
The author also address the broader question: are there fundamental reasons to prefer either ECN or delay for end-to-end congestion control in DC Networks ?
- Author: Radhika Mittal (UC Berkeley)
- Index: SIGCOMM 2015
- Reading date : 20/01/2019
- Keywords : Datacenter transport, delay-based CC, OS-bypass, RDMA
This is for [RDMA Section] paper 2. Aiming at reviewing the classical paper in relative to RDMA.
Datacenter transports aim to deliver low latency messageing together with high throughput. We show that simple packet delay, measured as round-trip times at hosts, is an effective congestion signal without the need for switch feedback.
- First, we show that advances in NIC hardware have made RTT measurement possible with microsecond accuracy, and that these RTTs are sufficient to estimate switch queueing.
- Then we describe how TIMELY can adjust transmission rates using RTT gradients to keep packet latency low while delivering high bandwidth.
TIMELY is the first delay-based CC protocol in datacenter. And it achieves its result despite having an order of magnitude fewer RTT signals.
- Author: Lisa Yan (Stanford University)
- Index: SIGCOMM 2017
- Reading date : 19/01/2019
- Keywords : Reproducible research, Teaching computer networks
- Score: ππππ
This paper is finished by my Academic Idol Lisa Yan. This paper is so important because Hope is a waking dreamοΌ
I'd like to deal with this course as a student.
I'd like to do the final pj of this course.
Lisa Yan said that:
"In the past five years, the graduate networking course at Stanford has assigned over 200 students the task of reproducing results from over 40 networking papers. We began the project as a means of teaching both engineering rigor and critical thinking, qualities that are necessary for careers in networking research and industry. We have observed that reproducing research can simultaneously be a tool for education and a means for students to contribute to the networking community. "
- Author: Yiming Kong (Georgia Tech)
- Index: SIGCOMM 2018 Workshop
- Reading date: 16/01/2019
- Keywords : Network-> Transport protocols Computing methodologiesβ Supervised learning; Reinforcement learning; Machine Learning; Security and Privacy.
This paper is about both off-line and on-line TCP-CC scheme.The writer use both supervised learning and reinforcement learning to train two TCP CC schemes.She use NS2 simulation to prepare for the dumbbell topology network and She set the environment to the basic under-buffered bottleneck network. Why i choose this paper to read in detail is because it goes into details about the TCP-CC-ML scheme And that helps me to playback the experiment in the paper.
- Author: Pavol Mulinka (CTU Czech)
- Index: SIGCOMM 2018 Workshop
- Reading date : 15/01/2019
- Keywords : Machine Learning; Computing methodologies; Security and Privacy.
This paper is about Data Stream Machine Learning for network security and anomaly detection. And especially for on-line questions (in this field the general approach is still off-line learning problem). The experiment result shows that the adaptive random forests and stochastic gradient descent models are able to keep up with important concept drifts by keeping high accuracy.
They also introduced Big-DAMA, a big data analytics framework for large scale network traffic monitoring and analysis.
- Author: Yibo Zhu (Microsoft;ByteDance)
- Index: SIGCOMM 2015
- Reading date : 02/01/2019-10/01/2019
- Keywords : Datacenter transport; RDMA; PFC; ECN; Congestion Control
This is for [RDMA Section] paper 1. Aiming at reviewing the classical paper in relative to RDMA. so i choose it due to its high reference in the filed of Datacenter transport.
This paper in one word, proposing a new way of congestion control to solve the head-of-line blocking and unfairness problem introduced by PFC. This way is very suitable for the RDMA Datacenter and works well with RoCEv2.
- Blog Conclusion of Congestion Control
- Reading date : 7/12/2018
- Keywords : Congestion Control ; Reinforcement Learning
I find a NYC coder's Blog in which it introduced several important concept in Congestion Control. Also with several useful code. So I read his 4 blogs and conclude his words. These 4 blogs are the introduction of CC, the Cubic , the RED , the Reinforcement Learning .
- Summary of Congestion Control
- Reading date : 5/12/2018
- Keywords : TCP ; Congestion Control
TCP Congestion Control aims at choose a strategy to find the appropriate cwnd and to acheive the best utility of bandwidth. In this summary, we briefly introduce several different mechanism in TCP CC Algorithm starting from TCP Vegas , TCP Reno , Cubic, BBR and Remy.
- Author: Ali Munir (Michigan State University)
- Index: SIGCOMM 2014
- Reading date : 3/12/2018
- Keywords : Datacenter transport; Congestion Control
Many Data Center transports have been proposed in recent times. Contrary to the common perception that they are competitors. We claim that the underlying strategies used in these protocols are , in fact complementary.
So we design PASE a transport framework that synthesize existing transport strategies, namely, self-adjusting endpoints (used in TCP style protocols), in-network prioritization (used in pFabric), and arbitration(used in PDQ). PASE is deployment friendly It does not require any changes to the network fabric. Its performance is better than state-of-the-art protocols (pFabric). Using NS2 simulation and testbed experiments.