Generating visual layouts is an essential ingredient of graphic design. The ability to condition layout generation on a partial subset of component attributes is critical to real-world applications that involve user interaction. Recently, diffusion models have demonstrated high-quality generative performances in various domains. However, it is unclear how to apply diffusion models to the natural representation of layouts which consists of a mix of discrete (class) and continuous (location, size) attributes. To address the conditioning layout generation problem, we introduce DLT, a joint discrete-continuous diffusion model. DLT is a transformer-based model which has a flexible conditioning mechanism that allows for conditioning on any given subset of all the layout component classes, locations, and sizes. Our method outperforms state-of-the-art generative models on various layout generation datasets with respect to different metrics and conditioning settings. Additionally, we validate the effectiveness of our proposed conditioning mechanism and the joint continuous-diffusion process. This joint process can be incorporated into a wide range of mixed discrete-continuous generative tasks. More information can be found on our project webpage: https://wix-incubator.github.io/DLT

Monitoring medical data, e.g., Electrocardiogram (ECG) signals, is a common application of Internet of Things (IoT) devices. Compression methods are often applied on the massive amounts of sensor data generated prior to sending it to the Cloud to reduce the storage and delivery costs. A lossy compression provides high compression gain (CG), but may reduce the performance of an ECG application (downstream task) due to information loss. Previous works on ECG monitoring focus either on optimizing the signal reconstruction or the task’s performance. Instead, we advocate a self-adapting lossy compression solution that allows configuring a desired performance level on the downstream tasks while maintaining an optimized CG that reduces Cloud costs.We propose Dynamic-Deep, a task-aware compression geared for IoT-Cloud architectures. Our compressor is trained to optimize the CG while maintaining the performance requirement of the downstream tasks chosen out of a wide range. In deployment, the IoT edge device adapts the compression and sends an optimized representation for each data segment, accounting for the downstream task’s desired performance without relying on feedback from the Cloud. We conduct an extensive evaluation of our approach on common ECG datasets using two popular ECG applications, which includes heart rate (HR) arrhythmia classification. We demonstrate that Dynamic-Deep can be configured to improve HR classification F1-score in a wide range of requirements. One of which is tuned to improve the F1-score by 3 and increases CG by up to 83% compared to the previous state-of-the-art (autoencoder-based) compressor. Analyzing Dynamic-Deep on the Google Cloud Platform, we observe a 97% reduction in cloud costs compared to a no compression solution.To the best of our knowledge, Dynamic-Deep is the first end-to-end system architecture proposal to focus on balancing the need for high performance of cloud-based downstream tasks and the desire to achieve optimized compression in IoT ECG monitoring settings.

Auto-scaling mechanisms are an important line of defense against Distributed Denial of Service (DDoS) in the cloud. Using auto-scaling, machines can be added and removed in an on-line manner to respond to fluctuating load. It is commonly believed that the auto-scaling mechanism casts DDoS attacks into Economic Denial of Sustainability (EDoS) attacks. Rather than suffering from performance degradation up to a total denial of service, the victim suffers only from the economic damage incurred by paying for the extra resources required to process the bogus traffic of the attack. Contrary to this belief, we present and analyze the Yo-Yo attack, a new attack against the auto-scaling mechanism, that can cause significant performance degradation in addition to economic damage. In the Yo-Yo attack, the attacker sends periodic bursts of overload, thus causing the auto-scaling mechanism to oscillate between scale-up and scale-down phases. The Yo-Yo attack is harder to detect and requires less resources from the attacker compared to traditional DDoS. We demonstrate the attack on Amazon EC2 [4], and analyze protection measures the victim can take by reconfiguring the auto-scaling mechanism.

We consider the problem of how to place and efficiently utilize resources in network environments. The setting consists of a regionally organized system which must satisfy regionally varying demands for various resources. The operator aims at placing resources in the regions as to minimize the cost of providing the demands. Examples of systems falling under this paradigm are 1) A peer supported Video on Demand service where the problem is how to place various video movies, and 2) A cloud-based system consisting of regional server-farms, where the problem is where to place various contents or end-user services. The main challenge posed by this paradigm is the need to deal with an arbitrary multi-dimensional (high-dimensionality) stochastic demand. We show that, despite this complexity, one can optimize the system operation while accounting for the full demand distribution. We provide algorithms for conducting this optimization and show that their complexity is pretty small, implying they can handle very large systems. The algorithms can be used for: 1) Exact system optimization, 2) deriving lower bounds for heuristic based analysis, and 3) Sensitivity analysis. The importance of the model is demonstrated by showing that an alternative analysis which is based on the demand means only, may, in certain cases, achieve performance that is drastically worse than the optimal one.

Clouds commonly store Virtual Machine (VM) images on networked storage. This poses a serious potential scalability bottleneck as launching a single fresh VM instance requires, at minimum, several hundred MB of network reads. As this bottleneck occurs most severely during read-intensive launching of new VMs, we focus on scalably minimizing time to boot a VM and load its critical applications.

Virtual Appliances (VAs) are Virtual Machines (VMs) geared towards a specific set of tasks. They require little or no configuration, working out-of-the-box. VAs fit neatly into the Cloud Computing paradigm - many copies of an identical machine can be launched in a data center, or home/business users can grab the appliance they need from the cloud to run locally just for so long as required. Companies and projects whose sole offerings are VAs ready for either desktop or data center use [3, 11] attest to the growing popularity of VAs. VMware's Appliance directory alone currently lists over 1400 VAs available for the VMware family of Virtual Machine Monitors (VMMs) [13].

TCP has traditionally been considered inappropriate for real-time applications. Nonetheless, popular applications such as Skype use TCP since UDP packets cannot pass through restrictive network address translators (NATs) and firewalls. Motivated by this observation, we study the delay performance of TCP for real-time media flows. We develop an analytical performance model for the delay of TCP. We use extensive experiments to validate the model and to evaluate the impact of various TCP mechanisms on its delay performance. Based on our results, we derive the working region for VoIP and live video streaming applications and provide guidelines for delay-friendly TCP settings. Our research indicates that simple application-level schemes, such as packet splitting and parallel connections, can reduce the delay of real-time TCP flows by as much as 30% and 90%, respectively.

TCP has traditionally been considered unfriendly for real-time applications. Nonetheless, popular applications such as Skype use TCP since UDP packets cannot pass through many NATs and firewalls. Motivated by this observation, we study the delay performance of TCP for real-time media flows. We develop an analytical performance model for the delay of TCP. We use extensive experiments to validate the model and to evaluate the impact of various TCP mechanisms on its delay performance. Based on our results, we derive the working region for VoIP and live video streaming applications and provide guidelines for delay-friendly TCP settings. Our research indicates that simple application-level schemes, such as packet splitting and parallel connections, can reduce the delay of real-time TCP flows by as much as 30% and 90%, respectively.

In this paper, we examine the feasibility of sending real-time CBR workloads over TCP. This is motivated by the friendliness of NATs and firewalls towards TCP as opposed to UDP as well as by recent improvements in Internet's bandwidth and loss rates. Traditionally, TCP has been considered undesirable for real-time CBR workloads. We evaluate this assertion by developing a novel analytical tool that yields TCP's sender-to-receiver socket delay distribution for CBR workloads. A key insight gained is that the use of smaller than MSS-sized packets in CBR workloads can exploit the TCP's ACK counting mechanism thereby limiting the delay impact of congestion window variations. We leverage this insight to provide a heuristic and system-level guidelines for reducing TCP transport delays.

In this paper we look at TCP data which was passively collected from an edge ISP, and analyze it to obtain some new results and deeper understanding of TCP loss process. The focus of our study is to identify the 'root cause' links, i.e., the links that are responsible for the majority of the losses or reorders found on the end-to-end TCP connection. We suggest a new root cause criterion and a cost-effective algorithm to identify the root cause links. The algorithm incorporates a new out-of-sequence packet classification technique which is interesting by itself. We test our algorithm on the collected and simulated data and analytically justify its correctness. The simulation results show that the algorithm has a 95% detection rate with 10% false detection rate. We also analyze TCP temporal loss process, and found that the burst loss size is geometrically distributed. We analyze the TCP time-out loss indication under the Bernoulli loss model, which is the simplest model that can cause a geometric distribution, and show that the behavior of the TCP loss process is not different than when tail drop is assumed.

We investigate the problem of finding the minimum delay application-layer multicast trees, such as the trees constructed in overlay networks. It is accepted that shortest path trees are not a good solution for the problem since such trees can have nodes with very large degree, termed high load nodes. The load on these nodes makes them a bottleneck in the distribution tree, due to computation load and access link bandwidth constrains. Many previous solutions limited the maximal degree of the nodes by introducing arbitrary constraints. In this work, we show how to directly map the node load to the delay penalty at the application host, and create a new model that captures the trade offs between the desire to select shortest path trees and the need to constrain the load on the hosts. In this model the problem is shown to be NP-hard. Therefore, we present a logarithmic approximation algorithm and an alternative heuristic solution. Our heuristic algorithm is shown by simulations to be scalable for large group sizes, and produces results that are very close to optimal.