Dynamic Graph Neural Networks (DGNNs) have been widely applied in various real-life applications, such as link prediction and pandemic forecast, to capture both static structural information temporal characteristics from dynamic graphs. Combining time-dependent -independent components, DGNNs manifest substantial parallel computation data reuse potentials, but suffer severe memory access inefficiency transfer overhead under the canonical one-graph-at-a-time training pattern. To tackle these...

With the increase in size and number of shared risk link groups (SRLGs) optical wavelength-division multiplexing (WDM) networks, capacity efficiency shared-path protection becomes much poorer due to SRLG-disjoint constraint, thus blocking probability higher. Furthermore, severe traps caused by SRLGs, it more difficult find an backup path with trap avoidance within reasonable computational complexity. As a result, mesh WDM network large SRLGs or SRLG size, 100% failure is no longer practical...

The Controller Area Network (CAN) has been widely used in the automotive and industrial automation for over two decades. However, due to lack of security mechanisms, CAN is vulnerable attacks. In this paper, we propose a novel protection scheme called CANeleon. It can defend against smart attacker who might inject malicious frames with legitimate frame IDs, which cannot be mitigated by existing countermeasures. Inspired idea moving target defense technologies, CANeleon equips each node...

Today, virtualization technique is increasingly mature and prevalent in server consolidation HPC. Virtual machine monitor plays a significant role the resource management by dynamically mapping virtual CPUs of machines to physical according chosen scheduling policy. However, since lacks insight into each machine, unpredictability workload makes effective allocation difficult. Particularly, current policy has critical impact on performance concurrent due non-synchronization CPUs. This paper...

As multi-core processors become increasingly mainstream, architects have likewise more interested in how best to make use of the computing capacity CPU, for instance, through multiple simultaneous threads or processes execution with OpenMP MPI. At same time, mature and prevailing virtualization technique server consolidation HPC promotes emergence a large number virtual SMP servers. Therefore, whether parallel program can run machine environment efficiently not is topic concern. In this...

GPUs have been becoming an indispensable computing platform in data centers, and co-locating multiple applications on the same GPU is widely used to improve resource utilization. However, performance interference due uncontrolled contention severely degrades of fails deliver satisfactory user experience. In this paper, we present SMGuard, a software approach flexibly manage usage under co-location. We also propose capacity based model CapSM, which provisions resources fine-grained...

The recent emergence of low-power high-throughput programmable storage platforms-SmartNIC JBOF (just-a-bunch-of-flash)-motivates us to rethink the cluster architecture and system stack for energy-efficient large-scale data-intensive workloads. Unlike conventional systems that use an array server JBOFs or embedded nodes, introduction SmartNIC has drastically changed compute, memory, I/O configurations. Such extremely imbalanced makes prior design philosophies techniques either ineffective invalid.

Memory access latency continues to be a dominant bottleneck in large class of applications on modern architectures. To optimize memory performance, it is important utilize the locality hierarchy. Data layout optimization can significantly improve locality. However, pinpointing inefficient code and providing insightful guidance for data challenging. Existing tools typically leverage heavyweight instrumentations, which hinders applicability these real long-running programs. address this issue,...

In a (t, n) threshold secret sharing scheme, any t- out-of-n participants could recover the shared secret, and less than t get nothing about secret. Most of existing schemes are not flexible enough for fixed threshold. this paper, new dynamic scheme was proposed, which is based on bilinear maps. The basic idea as follows: system consisted some dealer. Each participant holds only one permanent private key. dealer responsible to choose construct linear equations by using 'public keys. realized...

In this paper, we study the impact of shared risk link group (SRLG) failures on shared-path protection by examining percentage connections that are vulnerable after SRLG failures, investigate benefits backup reprovisioning and evaluate different policies for reprovisioning. Compared with single-link leave many more unprotected to next make network topology much sparser. The major challenge is how find SRLG-disjoint paths those a recovery ratio as high possible within reasonable computational...

Virtualization provides an effective management in server consolidation. The transparence enables different kinds of servers running the same platform, making full use hardware resource. However, virtualization introduces two-level schedulers: one from Guest OS, where tasks are scheduled to virtual CPUs (VCPUs), other machine monitor (VMM), VCPUs CPUs. As a result, lower level scheduler is ignorant task information so that it cannot allocate appropriate proportion CPU resource for every OS...

With DNN turning into the backbone of AI cloud services and propelling emergence INFerence-as-a-Service (INFaaS), DNN-specific accelerators have become indispensable components inference systems. Due to conservative "one-task-at-a-time" working mode deadline blindness those accelerators, implementing multi-tenancy that aims improve cost-effectiveness meet SLA requirements is intractable. Recent studies including temporal spatial approaches, employ manifold scheduling mechanisms sophisticated...

In the process of network operation, a large number alerts are generated every day, which reflect occurrence some abnormal conditions. Traditional methods depend too much on knowledge equipment manufacturers and industry experts, so we need novel ways to overcome this problem in management. The application data mining technology alarm pattern analysis has become focus current research. Researchers developed many kinds algorithms fitting different characteristics. This paper proposes concept...

Due to the broad attack surface and lack of runtime protection, potential safety security threats hinder real-life adoption autonomous vehicles. Although efforts have been made mitigate some specific attacks, there are few works on protection self-driving system. This paper presents a decentralized self-protection framework called Guardauto protect system against threats. First, proposes an isolation model decouple isolate its components with set partitions. Second, provides mechanisms for...

It is the requirement of next generation communication systems that devices should provide all existing and future services in integral way. After implementation 3G technology, smart now support text, voice video on device. With Wi-Fi technology low cost available market, users can enjoy even free with any one around world. If these technologies are brought together, result would be a large Distribute Wireless multimedia Network. This research work proposes such distributed wireless VoIP...

System dependability and real-time requirement have been receiving widespread attention in modern computer systems. Fault-tolerant scheduling provides a method of combining timing requirements faults tolerance for This paper presents preliminary study building fault-tolerant mechanism seL4 microkernel. seL4, as the first formally-verified operating system kernel world, is an ideal platform security-critical However, lack policy reduces its reliability user experience. We implemented basic...

Modern Graphic Processing Units (GPUs) have become pervasive computing devices in datacenters due to their high performance with massive thread level parallelism (TLP). GPUs are equipped large register files (RF) support fast context switch between threads and scratchpad memory (SPM) inter-thread communication within the cooperative array (CTA). However, TLP of is usually limited by inefficient resource management file memory. This inefficiency also leads underutilization. To overcome above...

Embedded virtualization systems become increasingly important in the light of mobile computing era. Current OKL4 technology has overcome some limitations plain by introducing microkernel approach but also imposes several concerns including incomplete IPC (Inter-Process Communication). With this mind, paper a novel asynchronous communication mechanism to optimize original is proposed. The able effectively deal with concurrent requests between any virtual machines (VMs) generating multiple...