197 skills found · Page 2 of 7
mbucchia / Quad Views FoveatedAn OpenXR API layer to emulate quad views and foveated rendering support on many headsets.
USTCPCS / CVPR2018 AttentionContext Encoding for Semantic Segmentation MegaDepth: Learning Single-View Depth Prediction from Internet Photos LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume On the Robustness of Semantic Segmentation Models to Adversarial Attacks SPLATNet: Sparse Lattice Networks for Point Cloud Processing Left-Right Comparative Recurrent Model for Stereo Matching Enhancing the Spatial Resolution of Stereo Images using a Parallax Prior Unsupervised CCA Discovering Point Lights with Intensity Distance Fields CBMV: A Coalesced Bidirectional Matching Volume for Disparity Estimation Learning a Discriminative Feature Network for Semantic Segmentation Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi- Supervised Semantic Segmentation Unsupervised Deep Generative Adversarial Hashing Network Monocular Relative Depth Perception with Web Stereo Data Supervision Single Image Reflection Separation with Perceptual Losses Zoom and Learn: Generalizing Deep Stereo Matching to Novel Domains EPINET: A Fully-Convolutional Neural Network for Light Field Depth Estimation by Using Epipolar Geometry FoldingNet: Interpretable Unsupervised Learning on 3D Point Clouds Decorrelated Batch Normalization Unsupervised Learning of Depth and Egomotion from Monocular Video Using 3D Geometric Constraints PU-Net: Point Cloud Upsampling Network Real-Time Monocular Depth Estimation using Synthetic Data with Domain Adaptation via Image Style Transfer Tell Me Where To Look: Guided Attention Inference Network Residual Dense Network for Image Super-Resolution Reflection Removal for Large-Scale 3D Point Clouds PlaneNet: Piece-wise Planar Reconstruction from a Single RGB Image Fully Convolutional Adaptation Networks for Semantic Segmentation CRRN: Multi-Scale Guided Concurrent Reflection Removal Network DenseASPP: Densely Connected Networks for Semantic Segmentation SGAN: An Alternative Training of Generative Adversarial Networks Multi-Agent Diverse Generative Adversarial Networks Robust Depth Estimation from Auto Bracketed Images AdaDepth: Unsupervised Content Congruent Adaptation for Depth Estimation DeepMVS: Learning Multi-View Stereopsis GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation Single-Image Depth Estimation Based on Fourier Domain Analysis Single View Stereo Matching Pyramid Stereo Matching Network A Unifying Contrast Maximization Framework for Event Cameras, with Applications to Motion, Depth, and Optical Flow Estimation Image Correction via Deep Reciprocating HDR Transformation Occlusion Aware Unsupervised Learning of Optical Flow PAD-Net: Multi-Tasks Guided Prediciton-and-Distillation Network for Simultaneous Depth Estimation and Scene Parsing Surface Networks Structured Attention Guided Convolutional Neural Fields for Monocular Depth Estimation TextureGAN: Controlling Deep Image Synthesis with Texture Patches Aperture Supervision for Monocular Depth Estimation Two-Stream Convolutional Networks for Dynamic Texture Synthesis Unsupervised Learning of Single View Depth Estimation and Visual Odometry with Deep Feature Reconstruction Left/Right Asymmetric Layer Skippable Networks Learning to See in the Dark
aicaprio / CurtainViewPull/Drawer/Layer View
nyorain / VilVulkan Layer for Live Introspection & Debugging. Allows to view all vulkan state live inside your application.
strongself / MRefreshThis pod enables you to add pull-to-refresh mechanism to your scrollviews and tableviews and use svg patterns in your refreshing view's layers.
cloud-native-toolkit / Multi Tenancy GitopsProvides our opinionated point of view on how GitOps can be used to manage the infrastructure, services and application layers of K8s based systems
pouyakary / TextGraphicTextGraphic is a framework for creating Textual Graphics. It provides layers, styling, rich color, text justification, layouts, tables, view-ports, transparency, etc.
SamHennessy / HliveHLive is a server-side WebSocket based dynamic template-less view layer for Go.
microsoft / Deep Language NetworksWe view Large Language Models as stochastic language layers in a network, where the learnable parameters are the natural language prompts at each layer. We stack two such layers, feeding the output of one layer to the next. We call the stacked architecture a Deep Language Network - DLN
google / Layered Scene InferenceCode accompanying the paper "Layer-structured 3D Scene Inference via View Synthesis", ECCV 2018
ryanswapp / React Meteor TemplateThis is a boilerplate template for a Meteor.js app that uses React.js as the view layer
Aryia-Behroziuan / NeuronsAn ANN is a model based on a collection of connected units or nodes called "artificial neurons", which loosely model the neurons in a biological brain. Each connection, like the synapses in a biological brain, can transmit information, a "signal", from one artificial neuron to another. An artificial neuron that receives a signal can process it and then signal additional artificial neurons connected to it. In common ANN implementations, the signal at a connection between artificial neurons is a real number, and the output of each artificial neuron is computed by some non-linear function of the sum of its inputs. The connections between artificial neurons are called "edges". Artificial neurons and edges typically have a weight that adjusts as learning proceeds. The weight increases or decreases the strength of the signal at a connection. Artificial neurons may have a threshold such that the signal is only sent if the aggregate signal crosses that threshold. Typically, artificial neurons are aggregated into layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first layer (the input layer) to the last layer (the output layer), possibly after traversing the layers multiple times. The original goal of the ANN approach was to solve problems in the same way that a human brain would. However, over time, attention moved to performing specific tasks, leading to deviations from biology. Artificial neural networks have been used on a variety of tasks, including computer vision, speech recognition, machine translation, social network filtering, playing board and video games and medical diagnosis. Deep learning consists of multiple hidden layers in an artificial neural network. This approach tries to model the way the human brain processes light and sound into vision and hearing. Some successful applications of deep learning are computer vision and speech recognition.[68] Decision trees Main article: Decision tree learning Decision tree learning uses a decision tree as a predictive model to go from observations about an item (represented in the branches) to conclusions about the item's target value (represented in the leaves). It is one of the predictive modeling approaches used in statistics, data mining, and machine learning. Tree models where the target variable can take a discrete set of values are called classification trees; in these tree structures, leaves represent class labels and branches represent conjunctions of features that lead to those class labels. Decision trees where the target variable can take continuous values (typically real numbers) are called regression trees. In decision analysis, a decision tree can be used to visually and explicitly represent decisions and decision making. In data mining, a decision tree describes data, but the resulting classification tree can be an input for decision making. Support vector machines Main article: Support vector machines Support vector machines (SVMs), also known as support vector networks, are a set of related supervised learning methods used for classification and regression. Given a set of training examples, each marked as belonging to one of two categories, an SVM training algorithm builds a model that predicts whether a new example falls into one category or the other.[69] An SVM training algorithm is a non-probabilistic, binary, linear classifier, although methods such as Platt scaling exist to use SVM in a probabilistic classification setting. In addition to performing linear classification, SVMs can efficiently perform a non-linear classification using what is called the kernel trick, implicitly mapping their inputs into high-dimensional feature spaces. Illustration of linear regression on a data set. Regression analysis Main article: Regression analysis Regression analysis encompasses a large variety of statistical methods to estimate the relationship between input variables and their associated features. Its most common form is linear regression, where a single line is drawn to best fit the given data according to a mathematical criterion such as ordinary least squares. The latter is often extended by regularization (mathematics) methods to mitigate overfitting and bias, as in ridge regression. When dealing with non-linear problems, go-to models include polynomial regression (for example, used for trendline fitting in Microsoft Excel[70]), logistic regression (often used in statistical classification) or even kernel regression, which introduces non-linearity by taking advantage of the kernel trick to implicitly map input variables to higher-dimensional space. Bayesian networks Main article: Bayesian network A simple Bayesian network. Rain influences whether the sprinkler is activated, and both rain and the sprinkler influence whether the grass is wet. A Bayesian network, belief network, or directed acyclic graphical model is a probabilistic graphical model that represents a set of random variables and their conditional independence with a directed acyclic graph (DAG). For example, a Bayesian network could represent the probabilistic relationships between diseases and symptoms. Given symptoms, the network can be used to compute the probabilities of the presence of various diseases. Efficient algorithms exist that perform inference and learning. Bayesian networks that model sequences of variables, like speech signals or protein sequences, are called dynamic Bayesian networks. Generalizations of Bayesian networks that can represent and solve decision problems under uncertainty are called influence diagrams. Genetic algorithms Main article: Genetic algorithm A genetic algorithm (GA) is a search algorithm and heuristic technique that mimics the process of natural selection, using methods such as mutation and crossover to generate new genotypes in the hope of finding good solutions to a given problem. In machine learning, genetic algorithms were used in the 1980s and 1990s.[71][72] Conversely, machine learning techniques have been used to improve the performance of genetic and evolutionary algorithms.[73] Training models Usually, machine learning models require a lot of data in order for them to perform well. Usually, when training a machine learning model, one needs to collect a large, representative sample of data from a training set. Data from the training set can be as varied as a corpus of text, a collection of images, and data collected from individual users of a service. Overfitting is something to watch out for when training a machine learning model. Federated learning Main article: Federated learning Federated learning is an adapted form of distributed artificial intelligence to training machine learning models that decentralizes the training process, allowing for users' privacy to be maintained by not needing to send their data to a centralized server. This also increases efficiency by decentralizing the training process to many devices. For example, Gboard uses federated machine learning to train search query prediction models on users' mobile phones without having to send individual searches back to Google.[74] Applications There are many applications for machine learning, including: Agriculture Anatomy Adaptive websites Affective computing Banking Bioinformatics Brain–machine interfaces Cheminformatics Citizen science Computer networks Computer vision Credit-card fraud detection Data quality DNA sequence classification Economics Financial market analysis[75] General game playing Handwriting recognition Information retrieval Insurance Internet fraud detection Linguistics Machine learning control Machine perception Machine translation Marketing Medical diagnosis Natural language processing Natural language understanding Online advertising Optimization Recommender systems Robot locomotion Search engines Sentiment analysis Sequence mining Software engineering Speech recognition Structural health monitoring Syntactic pattern recognition Telecommunication Theorem proving Time series forecasting User behavior analytics In 2006, the media-services provider Netflix held the first "Netflix Prize" competition to find a program to better predict user preferences and improve the accuracy of its existing Cinematch movie recommendation algorithm by at least 10%. A joint team made up of researchers from AT&T Labs-Research in collaboration with the teams Big Chaos and Pragmatic Theory built an ensemble model to win the Grand Prize in 2009 for $1 million.[76] Shortly after the prize was awarded, Netflix realized that viewers' ratings were not the best indicators of their viewing patterns ("everything is a recommendation") and they changed their recommendation engine accordingly.[77] In 2010 The Wall Street Journal wrote about the firm Rebellion Research and their use of machine learning to predict the financial crisis.[78] In 2012, co-founder of Sun Microsystems, Vinod Khosla, predicted that 80% of medical doctors' jobs would be lost in the next two decades to automated machine learning medical diagnostic software.[79] In 2014, it was reported that a machine learning algorithm had been applied in the field of art history to study fine art paintings and that it may have revealed previously unrecognized influences among artists.[80] In 2019 Springer Nature published the first research book created using machine learning.[81] Limitations Although machine learning has been transformative in some fields, machine-learning programs often fail to deliver expected results.[82][83][84] Reasons for this are numerous: lack of (suitable) data, lack of access to the data, data bias, privacy problems, badly chosen tasks and algorithms, wrong tools and people, lack of resources, and evaluation problems.[85] In 2018, a self-driving car from Uber failed to detect a pedestrian, who was killed after a collision.[86] Attempts to use machine learning in healthcare with the IBM Watson system failed to deliver even after years of time and billions of dollars invested.[87][88] Bias Main article: Algorithmic bias Machine learning approaches in particular can suffer from different data biases. A machine learning system trained on current customers only may not be able to predict the needs of new customer groups that are not represented in the training data. When trained on man-made data, machine learning is likely to pick up the same constitutional and unconscious biases already present in society.[89] Language models learned from data have been shown to contain human-like biases.[90][91] Machine learning systems used for criminal risk assessment have been found to be biased against black people.[92][93] In 2015, Google photos would often tag black people as gorillas,[94] and in 2018 this still was not well resolved, but Google reportedly was still using the workaround to remove all gorillas from the training data, and thus was not able to recognize real gorillas at all.[95] Similar issues with recognizing non-white people have been found in many other systems.[96] In 2016, Microsoft tested a chatbot that learned from Twitter, and it quickly picked up racist and sexist language.[97] Because of such challenges, the effective use of machine learning may take longer to be adopted in other domains.[98] Concern for fairness in machine learning, that is, reducing bias in machine learning and propelling its use for human good is increasingly expressed by artificial intelligence scientists, including Fei-Fei Li, who reminds engineers that "There’s nothing artificial about AI...It’s inspired by people, it’s created by people, and—most importantly—it impacts people. It is a powerful tool we are only just beginning to understand, and that is a profound responsibility.”[99] Model assessments Classification of machine learning models can be validated by accuracy estimation techniques like the holdout method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-cross-validation method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, bootstrap, which samples n instances with replacement from the dataset, can be used to assess model accuracy.[100] In addition to overall accuracy, investigators frequently report sensitivity and specificity meaning True Positive Rate (TPR) and True Negative Rate (TNR) respectively. Similarly, investigators sometimes report the false positive rate (FPR) as well as the false negative rate (FNR). However, these rates are ratios that fail to reveal their numerators and denominators. The total operating characteristic (TOC) is an effective method to express a model's diagnostic ability. TOC shows the numerators and denominators of the previously mentioned rates, thus TOC provides more information than the commonly used receiver operating characteristic (ROC) and ROC's associated area under the curve (AUC).[101] Ethics Machine learning poses a host of ethical questions. Systems which are trained on datasets collected with biases may exhibit these biases upon use (algorithmic bias), thus digitizing cultural prejudices.[102] For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants.[103][104] Responsible collection of data and documentation of algorithmic rules used by a system thus is a critical part of machine learning. Because human languages contain biases, machines trained on language corpora will necessarily also learn these biases.[105][106] Other forms of ethical challenges, not related to personal biases, are more seen in health care. There are concerns among health care professionals that these systems might not be designed in the public's interest but as income-generating machines. This is especially true in the United States where there is a long-standing ethical dilemma of improving health care, but also increasing profits. For example, the algorithms could be designed to provide patients with unnecessary tests or medication in which the algorithm's proprietary owners hold stakes. There is huge potential for machine learning in health care to provide professionals a great tool to diagnose, medicate, and even plan recovery paths for patients, but this will not happen until the personal biases mentioned previously, and these "greed" biases are addressed.[107] Hardware Since the 2010s, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks (a particular narrow subdomain of machine learning) that contain many layers of non-linear hidden units.[108] By 2019, graphic processing units (GPUs), often with AI-specific enhancements, had displaced CPUs as the dominant method of training large-scale commercial cloud AI.[109] OpenAI estimated the hardware compute used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017), and found a 300,000-fold increase in the amount of compute required, with a doubling-time trendline of 3.4 months.[110][111] Software Software suites containing a variety of machine learning algorithms include the following: Free and open-source so
laminas / Laminas ViewFlexible view layer supporting and providing multiple view layers, helpers, and more
ROGUE-JCTD / MapLoomMapLoom is a web mapping client based on OpenLayers 3. It provides full-featured editing capabilities for data stores in GeoServer - including PostGIS and GeoGig. For GeoGig layers, it lets you view history, and sync with other GeoGig repositories
royboy789 / Angular React Wordpress ThemeA WordPress theme built with AngularJS as the controller and ReactJS as the view layer. Powered by WP-API
ZhipingYang / VerticalTree🥶Provides a vertical drawing of the tree structure which can view information about the tree‘s nodes and supports console debug views & layers and so on
SOYJUN / FTP Implement Based On UDPThe aim of this assignment is to have you do UDP socket client / server programming with a focus on two broad aspects : Setting up the exchange between the client and server in a secure way despite the lack of a formal connection (as in TCP) between the two, so that ‘outsider’ UDP datagrams (broadcast, multicast, unicast - fortuitously or maliciously) cannot intrude on the communication. Introducing application-layer protocol data-transmission reliability, flow control and congestion control in the client and server using TCP-like ARQ sliding window mechanisms. The second item above is much more of a challenge to implement than the first, though neither is particularly trivial. But they are not tightly interdependent; each can be worked on separately at first and then integrated together at a later stage. Apart from the material in Chapters 8, 14 & 22 (especially Sections 22.5 - 22.7), and the experience you gained from the preceding assignment, you will also need to refer to the following : ioctl function (Chapter 17). get_ifi_info function (Section 17.6, Chapter 17). This function will be used by the server code to discover its node’s network interfaces so that it can bind all its interface IP addresses (see Section 22.6). ‘Race’ conditions (Section 20.5, Chapter 20) You also need a thorough understanding of how the TCP protocol implements reliable data transfer, flow control and congestion control. Chapters 17- 24 of TCP/IP Illustrated, Volume 1 by W. Richard Stevens gives a good overview of TCP. Though somewhat dated for some things (it was published in 1994), it remains, overall, a good basic reference. Overview This assignment asks you to implement a primitive file transfer protocol for Unix platforms, based on UDP, and with TCP-like reliability added to the transfer operation using timeouts and sliding-window mechanisms, and implementing flow and congestion control. The server is a concurrent server which can handle multiple clients simultaneously. A client gives the server the name of a file. The server forks off a child which reads directly from the file and transfers the contents over to the client using UDP datagrams. The client prints out the file contents as they come in, in order, with nothing missing and with no duplication of content, directly on to stdout (via the receiver sliding window, of course, but with no other intermediate buffering). The file to be transferred can be of arbitrary length, but its contents are always straightforward ascii text. As an aside let me mention that assuming the file contents ascii is not as restrictive as it sounds. We can always pretend, for example, that binary files are base64 encoded (“ASCII armor”). A real file transfer protocol would, of course, have to worry about transferring files between heterogeneous platforms with different file structure conventions and semantics. The sender would first have to transform the file into a platform-independent, protocol-defined, format (using, say, ASN.1, or some such standard), and the receiver would have to transform the received file into its platform’s native file format. This kind of thing can be fairly time consuming, and is certainly very tedious, to implement, with little educational value - it is not part of this assignment. Arguments for the server You should provide the server with an input file server.in from which it reads the following information, in the order shown, one item per line : Well-known port number for server. Maximum sending sliding-window size (in datagram units). You will not be handing in your server.in file. We shall create our own when we come to test your code. So it is important that you stick strictly to the file name and content conventions specified above. The same applies to the client.in input file below. Arguments for the client The client is to be provided with an input file client.in from which it reads the following information, in the order shown, one item per line : IP address of server (not the hostname). Well-known port number of server. filename to be transferred. Receiving sliding-window size (in datagram units). Random generator seed value. Probability p of datagram loss. This should be a real number in the range [ 0.0 , 1.0 ] (value 0.0 means no loss occurs; value 1.0 means all datagrams all lost). The mean µ, in milliseconds, for an exponential distribution controlling the rate at which the client reads received datagram payloads from its receive buffer. Operation Server starts up and reads its arguments from file server.in. As we shall see, when a client communicates with the server, the server will want to know what IP address that client is using to identify the server (i.e. , the destination IP address in the incoming datagram). Normally, this can be done relatively straightforwardly using the IP_RECVDESTADDR socket option, and picking up the information using the ancillary data (‘control information’) capability of the recvmsg function. Unfortunately, Solaris 2.10 does not support the IP_RECVDESTADDR option (nor, incidentally, does it support the msg_flags option in msghdr - see p.390). This considerably complicates things. In the absence of IP_RECVDESTADDR, what the server has to do as part of its initialization phase is to bind each IP address it has (and, simultaneously, its well-known port number, which it has read in from server.in) to a separate UDP socket. The code in Section 22.6, which uses the get_ifi_info function, shows you how to do that. However, there are important differences between that code and the version you want to implement. The code of Section 22.6 binds the IP addresses and forks off a child for each address that is bound to. We do not want to do that. Instead you should have an array of socket descriptors. For each IP address, create a new socket and bind the address (and well-known port number) to the socket without forking off child processes. Creating child processes comes later, when clients arrive. The code of Section 22.6 also attempts to bind broadcast addresses. We do not want to do this. It binds a wildcard IP address, which we certainly do not want to do either. We should bind strictly only unicast addresses (including the loopback address). The get_ifi_info function (which the code in Section 22.6 uses) has to be modified so that it also gets the network masks for the IP addresses of the node, and adds these to the information stored in the linked list of ifi_info structures (see Figure 17.5, p.471) it produces. As you go binding each IP address to a distinct socket, it will be useful for later processing to build your own array of structures, where a structure element records the following information for each socket : sockfd IP address bound to the socket network mask for the IP address subnet address (obtained by doing a bit-wise and between the IP address and its network mask) Report, in a ReadMe file which you hand in with your code, on the modifications you had to introduce to ensure that only unicast addresses are bound, and on your implementation of the array of structures described above. You should print out on stdout, with an appropriate message and appropriately formatted in dotted decimal notation, the IP address, network mask, and subnet address for each socket in your array of structures (you do not need to print the sockfd). The server now uses select to monitor the sockets it has created for incoming datagrams. When it returns from select, it must use recvfrom or recvmsg to read the incoming datagram (see 6. below). When a client starts, it first reads its arguments from the file client.in. The client checks if the server host is ‘local’ to its (extended) Ethernet. If so, all its communication to the server is to occur as MSG_DONTROUTE (or SO_DONTROUTE socket option). It determines if the server host is ‘local’ as follows. The first thing the client should do is to use the modified get_ifi_info function to obtain all of its IP addresses and associated network masks. Print out on stdout, in dotted decimal notation and with an appropriate message, the IP addresses and network masks obtained. In the following, IPserver designates the IP address the client will use to identify the server, and IPclient designates the IP address the client will choose to identify itself. The client checks whether the server is on the same host. If so, it should use the loopback address 127.0.0.1 for the server (i.e. , IPserver = 127.0.0.1). IPclient should also be set to the loopback address. Otherwise it proceeds as follows: IPserver is set to the IP address for the server in the client.in file. Given IPserver and the (unicast) IP addresses and network masks for the client returned by get_ifi_info in the linked list of ifi_info structures, you should be able to figure out if the server node is ‘local’ or not. This will be discussed in class; but let me just remind you here that you should use ‘longest prefix matching’ where applicable. If there are multiple client addresses, and the server host is ‘local’, the client chooses an IP address for itself, IPclient, which matches up as ‘local’ according to your examination above. If the server host is not ‘local’, then IPclient can be chosen arbitrarily. Print out on stdout the results of your examination, as to whether the server host is ‘local’ or not, as well as the IPclient and IPserver addresses selected. Note that this manner of determining whether the server is local or not is somewhat clumsy and ‘over-engineered’, and, as such, should be viewed more in the nature of a pedagogical exercise. Ideally, we would like to look up the server IP address(es) in the routing table (see Section 18.3). This requires that a routing socket be created, for which we need superuser privilege. Alternatively, we might want to dump out the routing table, using the sysctl function for example (see Section 18.4), and examine it directly. Unfortunately, Solaris 2.10 does not support sysctl. Furthermore, note that there is a slight problem with the address 130.245.1.123/24 assigned to compserv3 (see rightmost column of file hosts, and note that this particular compserv3 address “overlaps” with the 130.245.1.x/28 addresses in that same column assigned to compserv1, compserv2 & comserv4). In particular, if the client is running on compserv3 and the server on any of the other three compservs, and if that server node is also being identified to the client by its /28 (rather than its /24) address, then the client will get a “false positive” when it tests as to whether the server node is local or not. In other words, the client will deem the server node to be local, whereas in fact it should not be considered local. Because of this, it is perhaps best simply not to use compserv3 to run the client (but it is o.k. to use it to run the server). Finally, using MSG_DONTROUTE where possible would seem to gain us efficiency, in as much as the kernel does not need to consult the routing table for every datagram sent. But, in fact, that is not so. Recall that one effect of connect with UDP sockets is that routing information is obtained by the kernel at the time the connect is issued. That information is cached and used for subsequent sends from the connected socket (see p.255). The client now creates a UDP socket and calls bind on IPclient, with 0 as the port number. This will cause the kernel to bind an ephemeral port to the socket. After the bind, use the getsockname function (Section 4.10) to obtain IPclient and the ephemeral port number that has been assigned to the socket, and print that information out on stdout, with an appropriate message and appropriately formatted. The client connects its socket to IPserver and the well-known port number of the server. After the connect, use the getpeername function (Section 4.10) to obtain IPserver and the well-known port number of the server, and print that information out on stdout, with an appropriate message and appropriately formatted. The client sends a datagram to the server giving the filename for the transfer. This send needs to be backed up by a timeout in case the datagram is lost. Note that the incoming datagram from the client will be delivered to the server at the socket to which the destination IP address that the datagram is carrying has been bound. Thus, the server can obtain that address (it is, of course, IPserver) and thereby achieve what IP_RECVDESTADDR would have given us had it been available. Furthermore, the server process can obtain the IP address (this will, of course, be IPclient) and ephemeral port number of the client through the recvfrom or recvmsg functions. The server forks off a child process to handle the client. The server parent process goes back to the select to listen for new clients. Hereafter, and unless otherwise stated, whenever we refer to the ‘server’, we mean the server child process handling the client’s file transfer, not the server parent process. Typically, the first thing the server child would be expected to do is to close all sockets it ‘inherits’ from its parent. However, this is not the case with us. The server child does indeed close the sockets it inherited, but not the socket on which the client request arrived. It leaves that socket open for now. Call this socket the ‘listening’ socket. The server (child) then checks if the client host is local to its (extended) Ethernet. If so, all its communication to the client is to occur as MSG_DONTROUTE (or SO_DONTROUTE socket option). If IPserver (obtained in 5. above) is the loopback address, then we are done. Otherwise, the server has to proceed with the following step. Use the array of structures you built in 1. above, together with the addresses IPserver and IPclient to determine if the client is ‘local’. Print out on stdout the results of your examination, as to whether the client host is ‘local’ or not. The server (child) creates a UDP socket to handle file transfer to the client. Call this socket the ‘connection’ socket. It binds the socket to IPserver, with port number 0 so that its kernel assigns an ephemeral port. After the bind, use the getsockname function (Section 4.10) to obtain IPserver and the ephemeral port number that has been assigned to the socket, and print that information out on stdout, with an appropriate message and appropriately formatted. The server then connects this ‘connection’ socket to the client’s IPclient and ephemeral port number. The server now sends the client a datagram, in which it passes it the ephemeral port number of its ‘connection’ socket as the data payload of the datagram. This datagram is sent using the ‘listening’ socket inherited from its parent, otherwise the client (whose socket is connected to the server’s ‘listening’ socket at the latter’s well-known port number) will reject it. This datagram must be backed up by the ARQ mechanism, and retransmitted in the event of loss. Note that if this datagram is indeed lost, the client might well time out and retransmit its original request message (the one carrying the file name). In this event, you must somehow ensure that the parent server does not mistake this retransmitted request for a new client coming in, and spawn off yet another child to handle it. How do you do that? It is potentially more involved than it might seem. I will be discussing this in class, as well as ‘race’ conditions that could potentially arise, depending on how you code the mechanisms I present. When the client receives the datagram carrying the ephemeral port number of the server’s ‘connection’ socket, it reconnects its socket to the server’s ‘connection’ socket, using IPserver and the ephemeral port number received in the datagram (see p.254). It now uses this reconnected socket to send the server an acknowledgment. Note that this implies that, in the event of the server timing out, it should retransmit two copies of its ‘ephemeral port number’ message, one on its ‘listening’ socket and the other on its ‘connection’ socket (why?). When the server receives the acknowledgment, it closes the ‘listening’ socket it inherited from its parent. The server can now commence the file transfer through its ‘connection’ socket. The net effect of all these binds and connects at server and client is that no ‘outsider’ UDP datagram (broadcast, multicast, unicast - fortuitously or maliciously) can now intrude on the communication between server and client. Starting with the first datagram sent out, the client behaves as follows. Whenever a datagram arrives, or an ACK is about to be sent out (or, indeed, the initial datagram to the server giving the filename for the transfer), the client uses some random number generator function random() (initialized by the client.in argument value seed) to decide with probability p (another client.in argument value) if the datagram or ACK should be discarded by way of simulating transmission loss across the network. (I will briefly discuss in class how you do this.) Adding reliability to UDP The mechanisms you are to implement are based on TCP Reno. These include : Reliable data transmission using ARQ sliding-windows, with Fast Retransmit. Flow control via receiver window advertisements. Congestion control that implements : SlowStart Congestion Avoidance (‘Additive-Increase/Multiplicative Decrease’ – AIMD) Fast Recovery (but without the window-inflation aspect of Fast Recovery) Only some, and by no means all, of the details for these are covered below. The rest will be presented in class, especially those concerning flow control and TCP Reno’s congestion control mechanisms in general : Slow Start, Congestion Avoidance, Fast Retransmit and Fast Recovery. Implement a timeout mechanism on the sender (server) side. This is available to you from Stevens, Section 22.5 . Note, however, that you will need to modify the basic driving mechanism of Figure 22.7 appropriately since the situation at the sender side is not a repetitive cycle of send-receive, but rather a straightforward progression of send-send-send-send- . . . . . . . . . . . Also, modify the RTT and RTO mechanisms of Section 22.5 as specified below. I will be discussing the details of these modifications and the reasons for them in class. Modify function rtt_stop (Fig. 22.13) so that it uses integer arithmetic rather than floating point. This will entail your also having to modify some of the variable and function parameter declarations throughout Section 22.5 from float to int, as appropriate. In the unprrt.h header file (Fig. 22.10) set : RTT_RXTMIN to 1000 msec. (1 sec. instead of the current value 3 sec.) RTT_RXTMAX to 3000 msec. (3 sec. instead of the current value 60 sec.) RTT_MAXNREXMT to 12 (instead of the current value 3) In function rtt_timeout (Fig. 22.14), after doubling the RTO in line 86, pass its value through the function rtt_minmax of Fig. 22.11 (somewhat along the lines of what is done in line 77 of rtt_stop, Fig. 22.13). Finally, note that with the modification to integer calculation of the smoothed RTT and its variation, and given the small RTT values you will experience on the cs / sbpub network, these calculations should probably now be done on a millisecond or even microsecond scale (rather than in seconds, as is the case with Stevens’ code). Otherwise, small measured RTTs could show up as 0 on a scale of seconds, yielding a negative result when we subtract the smoothed RTT from the measured RTT (line 72 of rtt_stop, Fig. 22.13). Report the details of your modifications to the code of Section 22.5 in the ReadMe file which you hand in with your code. We need to have a sender sliding window mechanism for the retransmission of lost datagrams; and a receiver sliding window in order to ensure correct sequencing of received file contents, and some measure of flow control. You should implement something based on TCP Reno’s mechanisms, with cumulative acknowledgments, receiver window advertisements, and a congestion control mechanism I will explain in detail in class. For a reference on TCP’s mechanisms generally, see W. Richard Stevens, TCP/IP Illustrated, Volume 1 , especially Sections 20.2 - 20.4 of Chapter 20 , and Sections 21.1 - 21.8 of Chapter 21 . Bear in mind that our sequence numbers should count datagrams, not bytes as in TCP. Remember that the sender and receiver window sizes have to be set according to the argument values in client.in and server.in, respectively. Whenever the sender window becomes full and so ‘locks’, the server should print out a message to that effect on stdout. Similarly, whenever the receiver window ‘locks’, the client should print out a message on stdout. Be aware of the potential for deadlock when the receiver window ‘locks’. This situation is handled by having the receiver process send a duplicate ACK which acts as a window update when its window opens again (see Figure 20.3 and the discussion about it in TCP/IP Illustrated). However, this is not enough, because ACKs are not backed up by a timeout mechanism in the event they are lost. So we will also need to implement a persist timer driving window probes in the sender process (see Sections 22.1 & 22.2 in Chapter 22 of TCP/IP Illustrated). Note that you do not have to worry about the Silly Window Syndrome discussed in Section 22.3 of TCP/IP Illustrated since the receiver process consumes ‘full sized’ 512-byte messages from the receiver buffer (see 3. below). Report on the details of the ARQ mechanism you implemented in the ReadMe file you hand in. Indeed, you should report on all the TCP mechanisms you implemented in the ReadMe file, both the ones discussed here, and the ones I will be discussing in class. Make your datagram payload a fixed 512 bytes, inclusive of the file transfer protocol header (which must, at the very least, carry: the sequence number of the datagram; ACKs; and advertised window notifications). The client reads the file contents in its receive buffer and prints them out on stdout using a separate thread. This thread sits in a repetitive loop till all the file contents have been printed out, doing the following. It samples from an exponential distribution with mean µ milliseconds (read from the client.in file), sleeps for that number of milliseconds; wakes up to read and print all in-order file contents available in the receive buffer at that point; samples again from the exponential distribution; sleeps; and so on. The formula -1 × µ × ln( random( ) ) , where ln is the natural logarithm, yields variates from an exponential distribution with mean µ, based on the uniformly-distributed variates over ( 0 , 1 ) returned by random(). Note that you will need to implement some sort of mutual exclusion/semaphore mechanism on the client side so that the thread that sleeps and wakes up to consume from the receive buffer is not updating the state variables of the buffer at the same time as the main thread reading from the socket and depositing into the buffer is doing the same. Furthermore, we need to ensure that the main thread does not effectively monopolize the semaphore (and thus lock out for prolonged periods of time) the sleeping thread when the latter wakes up. See the textbook, Section 26.7, ‘Mutexes: Mutual Exclusion’, pp.697-701. You might also find Section 26.8, ‘Condition Variables’, pp.701-705, useful. You will need to devise some way by which the sender can notify the receiver when it has sent the last datagram of the file transfer, without the receiver mistaking that EOF marker as part of the file contents. (Also, note that the last data segment could be a “short” segment of less than 512 bytes – your client needs to be able to handle this correctly somehow.) When the sender receives an ACK for the last datagram of the transfer, the (child) server terminates. The parent server has to take care of cleaning up zombie children. Note that if we want a clean closing, the client process cannot simply terminate when the receiver ACKs the last datagram. This ACK could be lost, which would leave the (child) server process ‘hanging’, timing out, and retransmitting the last datagram. TCP attempts to deal with this problem by means of the TIME_WAIT state. You should have your receiver process behave similarly, sticking around in something akin to a TIME_WAIT state in case in case it needs to retransmit the ACK. In the ReadMe file you hand in, report on how you dealt with the issues raised here: sender notifying receiver of the last datagram, clean closing, and so on. Output Some of the output required from your program has been described in the section Operation above. I expect you to provide further output – clear, well-structured, well-laid-out, concise but sufficient and helpful – in the client and server windows by means of which we can trace the correct evolution of your TCP’s behaviour in all its intricacies : information (e.g., sequence number) on datagrams and acks sent and dropped, window advertisements, datagram retransmissions (and why : dup acks or RTO); entering/exiting Slow Start and Congestion Avoidance, ssthresh and cwnd values; sender and receiver windows locking/unlocking; etc., etc. . . . . The onus is on you to convince us that the TCP mechanisms you implemented are working correctly. Too many students do not put sufficient thought, creative imagination, time or effort into this. It is not the TA’s nor my responsibility to sit staring at an essentially blank screen, trying to summon up our paranormal psychology skills to figure out if your TCP implementation is really working correctly in all its very intricate aspects, simply because the transferred file seems to be printing o.k. in the client window. Nor is it our responsibility to strain our eyes and our patience wading through a mountain of obscure, ill-structured, hyper-messy, debugging-style output because, for example, your effort-conserving concept of what is ‘suitable’ is to dump your debugging output on us, relevant, irrelevant, and everything in between.
shapehq / SwiftDataDecoupledExample project showing how the data and view layers can be decoupled when using SwiftData for persistence.
nathanfaucett / Svelte Slateslate svelte view layer
bholmesdev / Astro On RailsRuby on Rails with Astro as the view layer
