Memory scalability has emerged as a critical bottleneck in virtualized cloud environments. Tiered memory architectures that combine limited fast memory with abundant slower memory offer a promising solution, but existing hypervisor-based approaches suffer from significant performance penalties. We present Demeter, introducing a paradigm shift through guest-delegated tiered memory management based on two key insights: (1) delegation to guests eliminates both expensive access tracking at the hypervisor level and frequent TLB flushes that severely degrade memory virtualization performance under two-dimensional address translation, and (2) Processor Event-Based Sampling, which cannot be effectively utilized by hypervisor-based solutions, remains fully functional and highly efficient when properly leveraged within the guest. Building on these insights, Demeter designs an efficient range-based tiered memory management scheme in guest virtual address space to preserve locality information and employs a double balloon-based provisioning mechanism that maintains cloud elasticity while enabling vendor-specific QoS control. Our evaluation with seven real-world workloads across DRAM+PMEM and DRAM+CXL.mem configurations demonstrates that Demeter improves performance by up to 2× compared to existing hypervisor-based approaches and by 28% on average compared to the next best guest-based alternative. Our implementation is fully open source and publicly available at Zenodo.

Disaggregated memory (DM) has garnered increasing attention due to high resource utilization. Fault tolerance is critical for key-value (KV) stores on DM since machine failures are common in datacenters. Existing KV stores on DM are generally based on replication to achieve fault tolerance, which however suffer from high memory space and performance overheads. In this paper, we investigate the efficiency of different fault-tolerant mechanisms on DM and reveal that checkpointing and erasure coding work best for the index and KV pairs respectively. Based on these observations, we present Aceso, a DM-based KV store that employs a hybrid fault-tolerant mechanism, combining checkpointing for the index and erasure coding for KV pairs. However, applying this hybrid mechanism to DM introduces multiple challenges, i.e., performance interference and data loss of checkpointing, slow space reclamation and failure recovery of erasure coding. To address these challenges, Aceso leverages a differential checkpointing scheme to reduce performance interference incurred by the bandwidth consumption to transmit checkpoints, a versioning approach to recover lost index updates on failures, a delta-based space reclamation mechanism to reclaim obsolete KV pairs with negligible overhead, and a tiered recovery scheme to minimize user disruption. Our experiments show that Aceso simultaneously achieves up to 2.7× throughput improvement, up to 54% tail latency reduction, and 44% memory space savings compared with the state-of-the-art replication-based KV store on DM.

Advanced Driver-Assistance Systems (ADAS) have been thriving and widely deployed in recent years. In general, these systems receive sensor data, compute driving decisions, and output control signals to the vehicles. To smooth out the uncertainties brought by sensor outputs, they usually leverage multi-sensor fusion (MSF) to fuse the sensor outputs and produce a more reliable understanding of the surroundings. However, MSF cannot completely eliminate the uncertainties since it lacks the knowledge about which sensor provides the most accurate data and how to optimally integrate the data provided by the sensors. As a result, critical consequences might happen unexpectedly. In this work, we observed that the popular MSF methods in an industry-grade ADAS can mislead the car control and result in serious safety hazards. We define the failures (e.g., car crashes) caused by the faulty MSF as fusion errors and develop a novel evolutionary-based domain-specific search framework, FusED, for the efficient detection of fusion errors. We further apply causality analysis to show that the found fusion errors are indeed caused by the MSF method. We evaluate our framework on two widely used MSF methods in two driving environments. Experimental results show that FusED identifies more than 150 fusion errors. Finally, we provide several suggestions to improve the MSF methods we study.

This paper proposes a co-design adaptive defense scheme against a class of zero-day buffer over-read attacks that follow unknown stationary probability distributions. In particular, the co-design scheme integrates an improved UCB algorithm and a customized server. The improved UCB algorithm adaptively allocates guard pages on a heap based on induced damage of the guard pages so as to minimize the accumulated damage over time. The security damages of the improved UCB algorithm are proven to be always below a temporal bound without knowing which attack is launched when the buffer allocation follows a certain stationary probability distribution. Then an efficient server modification is introduced to randomly allocate buffers. Moreover, the damages of our scheme asymptotically converge to those of the optimal defense policy where the launched attacks and their distributions are known in advance. Further, the co-design scheme is evaluated with several real-world Heartbleed attacks. The experiment results demonstrate the validity of the upper bound and show that the adaptive defense is effective against all the attacks of interest with runtime overheads as low as 5%.

DNN models have suffered from adversarial example attacks, which lead to inconsistent prediction results. As opposed to the gradient-based attack, which assumes white-box access to the model by the attacker, we focus on more realistic input perturbations from the real-world and their actual impact on the model robustness without any presence of the attackers. In this work, we promote a standardized framework to quantify the robustness against real-world threats. It is composed of a set of safety properties associated with common violations, a group of metrics to measure the minimal perturbation that causes the offense, and various criteria that reflect different aspects of the model robustness. By revealing comparison results through this framework among 13 pre-trained ImageNet classifiers, three state-of-the-art object detectors, and three cloud-based content moderators, we deliver the status quo of the real-world model robustness. Beyond that, we provide robustness benchmarking datasets for the community.

Return-oriented programming (ROP) is a code reuse attack that chains short snippets of existing code to perform arbitrary operations on target machines. Existing detection methods against ROP exhibit unsatisfactory detection accuracy and/or have high runtime overhead. In this paper, we present DeepReturn, which innovatively combines address space layout guided disassembly and deep neural networks to detect ROP payloads. The disassembler treats application input data as code pointers and aims to find any potential gadget chains, which are then classified by a deep neural network as benign or malicious. Our experiments show that DeepReturn has high detection rate (99.3%) and a very low false positive rate (0.01%). DeepReturn successfully detects all of the 100 real-world ROP exploits that are collected in-the-wild, created manually or created by ROP exploit generation tools. DeepReturn is non-intrusive and does not incur any runtime overhead to the protected program.

Growing multi-stage attacks in computer networks impose significant security risks and necessitate the development of effective defense schemes that are able to autonomously respond to intrusions during vulnerability windows. However, the defender faces several real-world challenges, e.g., unknown likelihoods and unknown impacts of successful exploits. In this article, we leverage reinforcement learning to develop an innovative adaptive cyber defense to maximize the cost-effectiveness subject to the aforementioned challenges. In particular, we use Bayesian attack graphs to model the interactions between the attacker and networks. Then we formulate the defense problem of interest as a partially observable Markov decision process problem where the defender maintains belief states to estimate system states, leverages Thompson sampling to estimate transition probabilities, and utilizes reinforcement learning to choose optimal defense actions using measured utility values. The algorithm performance is verified via numerical simulations based on real-world attacks.

In the wake of the research community gaining deep understanding about control-hijacking attacks, data-oriented attacks have emerged. Among data-oriented attacks, data structure manipulation attack (DSMA) is a major category. Pioneering research was conducted and shows that DSMA is able to circumvent the most effective defenses against control-hijacking attacks — DEP, ASLR and CFI. Up to this day, only two defense techniques have demonstrated their effectiveness: Data Flow Integrity (DFI) and Data Structure Layout Randomization (DSLR). However, DFI has high performance overhead, and dynamic DSLR has two main limitations. L-1: Randomizing a large set of data structures will significantly affect the performance. L-2: To be practical, only a fixed sub-set of data structures are randomized. In the case that the data structures targeted by an attack are not covered, dynamic DSLR is essentially noneffective. To address these two limitations, we propose a novel technique, feedback-control-based adaptive DSLR and build a system named SALADSPlus. SALADSPlus seeks to optimize the trade-off between security and cost through feedback control. Using a novel feedback-control-based adaptive algorithm extended from the Upper Confidence Bound (UCB) algorithm, the defender (controller) uses the feedbacks (cost-effectiveness) from previous randomization cycles to adaptively choose the set of data structures to randomize (the next action). Different from dynamic DSLR, the set of randomized data structures are adaptively changed based on the feedbacks. To obtain the feedbacks, SALADSPlus inserts canary in each data structure at the time of compilation. We have implemented SALADSPlus based on gcc-4.5.0. Experimental results show that the runtime overheads are 1.8%, 3.7%, and 5.3% when the randomization cycles are selected as 10s, 5s, and 1s respectively.

Address space randomization has long been used for counteracting code reuse attacks, ranging from conventional ROP to sophisticated Just-in-Time ROP. At the high level, it shuffles program code in memory and thus prevents malicious ROP payload from performing arbitrary operations. While effective in mitigating attacks, existing randomization mechanisms are impractical for real-world applications and systems, especially considering the significant performance overhead and potential program corruption incurred by their implementation. In this paper, we introduce CHAMELEON, a practical defense mechanism that hinders code reuse attacks, particularly Just-in-Time ROP attacks. Technically speaking, CHAMELEON instruments program code, randomly shuffles code page addresses and minimizes the attack surface exposed to adversaries. While this defense mechanism follows in the footprints of address space randomization, our design principle focuses on using randomization to obstruct code page disclosure, making the ensuing attacks infeasible. We implemented a prototype of CHAMELEON on Linux operating system and extensively experimented it in different settings. Our theoretical and empirical evaluation indicates the effectiveness and efficiency of CHAMELEON in thwarting Just-in-Time ROP attacks.