We present \sys, a declarative, unsupervised root-cause analysis engine that uses time series monitoring data from large complex systems such as data centres. \sys empowers operators to succinctly specify a large number of causal hypotheses to search for causes of interesting events. \sys then ranks these hypotheses, reducing the number of causal dependencies from hundreds of thousands to a handful for human understanding. We show how a declarative language, such as SQL, can be effective in declaratively enumerating hypotheses that probe the structure of an unknown probabilistic graphical causal model of the underlying system. Our thesis is that databases are in a unique position to enable users to rapidly explore the possible causal mechanisms in data collected from diverse sources. We empirically demonstrate how \sys had helped us resolve over 30~performance issues in a commercial product since late 2014, of which we discuss a few cases in detail.

The application of congestion control can have a significant detriment to the quality of service experienced at higher layers, especially under high packet loss rates. The effects of throughput loss due to the congestion control misinterpreting packet losses in poor channels is further compounded for applications such as HTTP and video leading to a significant decrease in the user's quality of service. Therefore, we consider the application of congestion control to transport layer packet streams that use error-correction coding in order to recover from packet losses. We introduce a modified AIMD approach, develop an approximate mathematic model suited to performance analysis, and present extensive experimental measurements in both the lab and the “wild” to evaluate performance. Our measurements highlight the potential for remarkable performance gains, in terms of throughput and upper layer quality of service, when using coded transports.

We take an analytical approach to study fundamental rate-delay-reliability trade-offs in the context of media streaming. We consider the probability of interruption in media playback (buffer underflow) as well as the number of initially buffered packets (initial waiting time) as the Quality of user Experience (QoE) metrics. We characterize the optimal trade-off between these metrics as a function of system parameters such as the packet arrival rate and file size, for different channel models. In the first model, we assume packets arrive according to independent Poisson processes from multiple servers or peers. We use random linear network coding to simplify the packet requests at the network layer and avoid duplicate packet reception. This allows us to model the receiver's buffer as a queue with Poisson arrivals and deterministic departures. For this model, we show that for arrival rates slightly larger than the play rate, the minimum initial buffering required to achieve certain level of interruption probability remains bounded as the file size grows. This is not the case when the arrival rate and the play rate match. In the second model, we consider channels with memory, which can be modeled using Markovian arrival processes. We characterize the optimal trade-off curves for the infinite file size case, in such Markovian environments.

We take an analytical approach to study Quality of user Experience (QoE) for media streaming applications. We use the fact that random linear network coding applied to blocks of video frames can significantly simplify the packet requests at the network layer and avoid duplicate packet reception. We model the receiver's buffer as a queue with Poisson arrivals and deterministic departures. We consider the probability of interruption in video playback (buffer underflow) as well as the number of initially buffered packets (initial waiting time) as the QoE metrics. We explicitly characterize the optimal trade-off between these metrics by providing upper and lower bounds on the minimum initial buffering required to achieve certain level of interruption probability for different regimes of the system parameters. Our bounds are asymptotically tight as the file size goes to infinity. Further, we show that for arrival rates slightly larger than the play rate, the minimum initial buffering remains bounded as the file size grows. This is not the case when the arrival rate and the play rate match.

In this paper, we consider the problem of rate and power allocation in a multiple-access channel (MAC). Our objective is to obtain rate and power allocation policies that maximize a general concave utility function of average transmission rates on the information-theoretic capacity region of the MAC without using queue-length information. First, we address the utility maximization problem in a nonfading channel and present a gradient projection algorithm with approximate projections. By exploiting the polymatroid structure of the capacity region, we show that the approximate projection can be implemented in time polynomial in the number of users. Second, we present optimal rate and power allocation policies in a fading channel where channel statistics are known. For the case that channel statistics are unknown and the transmission power is fixed, we propose a greedy rate allocation policy and characterize the performance difference of this policy and the optimal policy in terms of channel variations and structure of the utility function. The numerical results demonstrate superior convergence rate performance for the greedy policy compared to queue-length-based policies. In order to reduce the computational complexity of the greedy policy, we present approximate rate allocation policies which track the greedy policy within a certain neighborhood.

We consider the problem of rate allocation in a fading Gaussian multiple-access channel with fixed transmission powers. The goal is to maximize a general concave utility function of the expected achieved rates of the users. There are different approaches to this problem in the literature. From an information theoretic point of view, rates are allocated only by using the channel state information. The queueing theory approach utilizes the global queue-length information for rate allocation to guarantee throughput optimality as well as maximizing a utility function of the rates. In this work, we make a connection between these two approaches by showing that the information theoretic capacity region of a multiple-access channel and its stability region are equivalent. Moreover, our numerical results show that a simple greedy policy which does not use the queue-length information can outperform queue-length based policies in terms of convergence rate and fairness.