Asymmetric Workloads Research Articles

Multiplexed Asymmetric Attacks: Next-Generation DDoS on HTTP/2 Servers

Distributed Denial of Service (DDoS) attacks using the HTTP protocol have started gaining popularity in recent years. A recent trend in this direction has been the use of computationally expensive requests to launch attacks. These attacks, called Asymmetric Workload attacks can bring down servers using limited resources, and are extremely difficult to detect. The introduction of HTTP/2 has been welcomed by developers because it improves user experience and efficiency. This was made possible by the ability to transport HTTP requests and their associated inline resources simultaneously by using Multiplexing and Server Push. However multiplexing has made request traffic bursty and rendered DDoS detection mechanisms based on connection limiting obsolete. Contrary to its intention, multiplexing can also be misused to launch sophisticated DDoS attacks using multiple high workload requests in a single TCP connection. However, sufficient research has not been done in this area. Existing research demonstrates that the HTTP/2 protocol allows users to launch DDoS attacks easily, but does not focus on whether an HTTP/2 server can handle DDoS attacks more efficiently or not. Also, sufficient research has not been done on the possibility of Multiplexing and Server Push being misused. In this work, we analyse the performance of an HTTP/2 server compared to an HTTP/1.1 server under an Asymmetric DDoS attack for the same load. We propose a new DDoS attack vector called a Multiplexed Asymmetric DDoS attack, which uses multiplexing in a different way than intended. We show that such an attack can bring down a server with just a few attacking clients. We also show that a Multiplexed Asymmetric Attack on a server with Server Push enabled can trigger an egress network layer flood in addition to an application layer attack.

CUBE: A scalable framework for large-scale industrial simulations

Writing high-performance solvers for engineering applications is a delicate task. These codes are often developed on an application to application basis, highly optimized to solve a certain problem. Here, we present our work on developing a general simulation framework for efficient computation of time-resolved approximations of complex industrial flow problems—Complex Unified Building cube method (CUBE). To address the challenges of emerging, modern supercomputers, suitable data structures and communication patterns are developed and incorporated into CUBE. We use a Cartesian grid together with various immersed boundary (IB) methods to accurately capture moving, complex geometries. The asymmetric workload of the IB is balanced by a predictive dynamic load balancer, and a multithreaded halo exchange algorithm is employed to efficiently overlap communication with computations. Our work also concerns efficient methods for handling the large amount of data produced by large-scale flow simulations, such as scalable parallel I/O, data compression, and in-situ processing.

SoMed: scaling decentralised microblogging services with a hybrid DHT framework

Today's microblogging services such as Twitter have experienced unparalleled growth. The microblogging services have served more as news media outlets than online social networks. We use them to participate in some big events, as exemplified by predicting the H1N1 virus emergency, detecting earthquakes, and living World Cup Games. However, these kinds of burst events will cause peak access. It is unsurprising that servers will suffer from large amount of concurrent access and lead to significant impairment of availability. Furthermore, it will lead to asymmetric workloads, because of the divergent behaviours of users, social users and media users. Some media users or celebrities e.g. Lady Gaga accounts for over 20 million followers will generate a huge amount of traffic and lead to tremendous workloads. In this paper, we are in consideration of these problems together, and present SoMed, a hybrid DHT framework towards decentralised microblogging services. We evaluate SoMed by conducting simulations with real traces of Twitter. An experimental analysis illustrates significantly bandwidth saving and response time reducing with high QoS.

A 2D algorithm with asymmetric workload for the UPC conjugate gradient method

This paper examines four different strategies, each one with its own data distribution, for implementing the parallel conjugate gradient (CG) method and how they impact communication and overall performance. Firstly, typical 1D and 2D distributions of the matrix involved in CG computations are considered. Then, a new 2D version of the CG method with asymmetric workload, based on leaving some threads idle during part of the computation to reduce communication, is proposed. The four strategies are independent of sparse storage schemes and are implemented using Unified Parallel C (UPC), a Partitioned Global Address Space (PGAS) language. The strategies are evaluated on two different platforms through a set of matrices that exhibit distinct sparse patterns, demonstrating that our asymmetric proposal outperforms the others except for one matrix on one platform.

On the impact of serializing contention management on STM performance

Transactional memory (TM) is an emerging concurrent programming abstraction. Numerous software-based transactional memory (STM) implementations have been developed in recent years. STM implementations must guarantee transaction atomicity and isolation. In order to ensure progress, an STM implementation must resolve transaction collisions by consulting a contention manager (CM).Recent work established that serializing contention management–a technique in which the execution of colliding transactions is serialized for eliminating repeat-collisions–can dramatically improve STM performance in high-contention workloads. In low-contention and highly-parallel workloads, however, excessive serialization of memory transactions may limit concurrency too much and hurt performance. It is therefore important to better understand how the impact of serialization on STM performance varies as a function of workload characteristics.We investigate how serializing CM influences the performance of STM systems. Specifically, we study serialization’s influence on STM throughput (number of committed transactions per time unit) and efficiency (ratio between the extent of “useful” work done by the STM and work “wasted” by aborts) as the workload’s level of contention changes. Towards this goal, we implement CBench—a synthetic benchmark that generates workloads in which transactions have (parameter) pre-determined length and probability of being aborted in the lack of contention reduction mechanisms. CBench facilitates evaluating the efficiency of contention management algorithms across the full spectrum of contention levels.The characteristics of TM workloads generated by real applications may vary over time. To achieve good performance, CM algorithms need to monitor these characteristics and change their behavior accordingly. We implement adaptive algorithms that control the activation of serializing CM according to measured contention level, based on a novel low-overhead serialization mechanism. We then evaluate our new algorithms on CBench-generated workloads and on additional well-known STM benchmark applications.Our results shed light on the manner in which serializing CM should be used by STM systems. We show that adaptive contention managers are susceptible to a phenomenon of mode oscillations–in which serialization is repeatedly turned on and off–which hurts performance. We implement a simple stabilizing mechanism that solves this problem. We also compare the performance of local and global adaptive CM algorithms and demonstrate that local adaptive algorithms are superior for applications with asymmetric workloads.

An analytic study of dynamic hardware and software cache coherence strategies

Dynamic software cache coherence strategies use information about program sharing behaviour to manage caches at run-time and at a granularity defined by the application. The program-level information is obtained through annotations placed into the application by the user or the compiler. The coherence protocols may range from simple static algorithms to dynamic algorithms that use run-time data structures similar to the directories used in hardware strategies. In this paper, we present an analytic study of five dynamic software cache coherence algorithms and compare these to a representative hardware coherence strategy. The analytic model is constructed using four input parameters --- write probability, locality, granularity, and system size --- and solved by analysis of a Markov chain. We show that the fundamental tradeoffs between the different hardware and software strategies are captured in this model. The results of the study show that hardware schemes perform better for fine-grained data structures for much of the parameter space that we study. However, for coarse-grained data structures, various software algorithms are dominant over most of the parameter space. Further, hardware strategies are found to be more susceptible to the effects of contention, and also perform worse for the asymmetric workload that we study.

Parallel Discrete Event Simulation on SIMD Computers

A simulator based on a synchronous parallel simulation scheme is developed on an SIMD computer, a connection machine (CM-2) with 8K processors. Multistage interconnection networks of different sizes (2 to 16 stages) are simulated. Two categories of experiments are performed: symmetric and asymmetric workloads. Four metrics are used to evaluate performance of the simulator: degree of detected parallelism, degree of potential parallelism, speedup, and efficiency. The obtained results demonstrate good speedups (over 2000) in some cases.

Decomposition approach to evaluate CSMA/CD protocol performance with asymmetric workload

A model is proposed for the analysis of a p-persistent, slotted CSMA/CD protocol on a local area network consisting of a finite population of buffered stations with different workloads. The analysis of the protocol is performed using a decomposition strategy, and considering each station as an M/G/1 queueing system. Since stations interact with each other in channel contention, the interference is modelled through the service time probability-density-function L-Transform of the queueing system. The average packet delay characteristics and the stability station behaviour are obtained; to prove the model accuracy, numerical results from both analysis and simulation are given. The influence of scheduling probability p on system stability and average packet delay are analysed. Furthermore, an assignment strategy for probability p is proposed to improve system performance with asymmetric workloads. Finally, the analysis is extended to the transmission of packet bursts on the channel.