Pythia: Scheduling of Concurrent Network Packet Processing Applications on Heterogeneous Devices

Abstract

Modern commodity computing systems are composed of a number of heterogeneous processing units, each one with its own unique performance and energy characteristics. However, the majority of current network packet processing frameworks targets only one device (either the CPU or an accelerator), leaving the remaining computational resources underutilized or even idle. In this paper, we propose an adaptive scheduling approach for network packet processing applications that exploits any heterogeneous architecture that can be found in a commodity high-end hardware setup. Our scheduler not only distributes the workloads to the appropriate devices in the system to achieve the desired performance results, but also enables the multiplexing of diverse, concurrently executed network packet processing applications, eliminating the interference effects introduced at run-time. The evaluation results show that our scheduler is able to tackle any interference in the shared hardware resources as well to respond quickly to dynamic fluctuations (e.g., application overloads, traffic bursts, infrastructural changes, etc.) that may occur at real time.

The Best of Many Worlds: Scheduling Machine Learning Inference on CPU-GPU Integrated Architectures

“…b) Heterogeneous processing environments. : The architectural heterogeneity in commodity computing systems has led many researchers to explore their abilities on different applications and workloads [19], [20], [21], [22], [23], [24], [25], [26], [27]. The performance ranking of different devices has been shown to have wide variations when executing different classes of network applications [19], [21], [22].…”

“…: The architectural heterogeneity in commodity computing systems has led many researchers to explore their abilities on different applications and workloads [19], [20], [21], [22], [23], [24], [25], [26], [27]. The performance ranking of different devices has been shown to have wide variations when executing different classes of network applications [19], [21], [22]. To capture the performance variability effect, it is important to perform in-field studies and quantitative evaluation of the different processing units [26], [27].…”

The Diversification and Enhancement of an IDS Scheme for the Cybersecurity Needs of Modern Supply Chains

“…Furthermore, Papadogiannaki et al [30] proposed a scheduling approach that, based on performance policies (such as high throughput or low power consumption), determines the most suitable combination of heterogeneous devices (i.e., CPU, integrated, or discrete GPUs) for efficient execution of network packet processing workloads (such as DPI or packet encryption). Similarly, in Pythia [31,32], authors add the support for concurrent execution of different network packet processing applications across multiple and heterogeneous devices. In APUNet [33], authors propose the utilization of integrated GPUs to accelerate packet processing workloads without paying the overheads of memory transactions between the host and discrete GPUs.…”

A Survey on Encrypted Network Traffic Analysis Applications, Techniques, and Countermeasures

“…Popular network intrusion detection system (NIDS) solutions like Snort [22] and Suricata [21] utilize pattern matching and regular expressions in order to analyze network traffic while Zeek/Bro [23] utilizes scripts that allow easier automation. The research community has also put effort into improving the performance of NIDS using either commodity hardware, such as graphics processing units (GPUs) [69,70,76,116,138,[152][153][154][155] and parallel nodes [117,149], or specialized hardware, such as ternary content-addressable memories (TCAMs), application-specific integrated circuits (ASICs), and field-programmable gate array (FPGAs) [47,99,141,169]. However, these works are able to process network traffic that is unencrypted, since they extract meaningful information from network packet payload content.…”