DeepMatch

Hypolite, Joel; Sonchack, John; Hershkop, Shlomo; Dautenhahn, Nathan; DeHon, André; Smith, Jonathan M.

doi:10.1145/3386367.3431290

Cited by 32 publications

(3 citation statements)

References 56 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Early SmartNICs [8], [9], [10] provided a limited set of in-pipeline processing capabilities for delivering packets to user space [11], traffic shaping [12], or tailored offloads for TCP packet processing [13]. Emerging SmartNICs provide much richer hardware capabilities such as regular expression matching [14], compression [15], encryption [16], and SDN controllers [17]. Current commercial SmartNIC products exist at multiple points along this design space and range from generally programmable in-pipeline processing capabilities [18], [19], SmartNICs that provide in-pipeline processing with general programs and predefined hardware functions [20], and the most capable SmartNICs providing a complete feature set that enables packet processing both alongside and within the packet pipeline using both predefined programs and programmable processors [5], [21].…”

Section: A From Smartnics To Dpusmentioning

confidence: 99%

Optimizing Application Performance with BlueField: Accelerating Large-Message Blocking and Nonblocking Collective Operations

Graham,

Bosilca,

Qin

et al. 2024

ISC High Performance 2024 Research Paper Proceedings (39th International Conference)

View full text Add to dashboard Cite

With the end of Dennard scaling, specializing and distributing compute engines throughout the system is a promising technique to improve applications performance. For example, NVIDIA's BlueField Data Processing Unit (DPU) integrates programmable processing elements within the network and offers specialized network processing capabilities. These capabilities enable communication via offloads onto DPUs and present new application opportunities for offloading nonblocking or complex communication patterns such as collective communication operations. This paper discusses the lessons learned enabling DPUbased acceleration for collective communication algorithms by describing the impact of such offloaded collective operations on two applications: Octopus and P3DFFT++. We present new algorithms for the nonblocking MPI Ialltoallv and blocking MPI Allgatherv collective operations that leverage DPU offloading, which are used by the above applications, and evaluate them. Our experiments show a performance improvement in the range of 14% to 49% for P3DFFT++ and 17% for Octopus, even though the performance of those collectives in well-balanced OSU latency benchmarks shows comparable performance to welloptimized host-based implementations of these collectives. This demonstrates that taking into account load imbalance in communication algorithms can help improve application performance where such imbalance is common and large in magnitude.

show abstract

Section: A From Smartnics To Dpusmentioning

confidence: 99%

Optimizing Application Performance with BlueField: Accelerating Large-Message Blocking and Nonblocking Collective Operations

Graham,

Bosilca,

Qin

et al. 2024

ISC High Performance 2024 Research Paper Proceedings (39th International Conference)

View full text Add to dashboard Cite

show abstract

“…DeepMatch [15] provides a line-rate DPI primitive in the data plane using Netronome SMART. DeepMatch showcases new applications in the data plane that were not realized before, such as Quality of Service (QoS) policies and network monitoring that require processing application layer information.…”

Section: B Deep Packet Inspection (Dpi) In P4mentioning

confidence: 99%

P4DDPI: Securing P4-Programmable Data Plane Networks via DNS Deep Packet Inspection

AlSabeh¹,

Kfoury²,

Crichigno³

et al. 2022

Proceedings 2022 Workshop on Measurements, Attacks, and Defenses for the Web

View full text Add to dashboard Cite

“…However, with the increasing use of advanced encryption, traditional DPI tools struggle to meet the latest network traffic needs and requirements. This has resulted in the necessity for more advanced DPI methods capable of processing encrypted traffic using machine learning (ML) algorithms [8]. Similarly, machine learning (ML) is built on the concept of features, which are parameters within a dataset.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning-based Enhanced Deep Packet Inspection for IP Packet Priority Classification with Differentiated Services Code Point for Advance Network Management

Khan,

Ibrahim

2024

JTEC

View full text Add to dashboard Cite

In modern networking, the efficient prioritization and classification of network traffic is paramount to ensure optimal network performance and optimization. This study presents an approach to enhance intelligent packet forwarding priority classification on Differentiated Services Code Point (DSCP), leveraging classifiers from machine learning algorithms for Deep Packet Inspection (DPI). The DSCP resides inside the Differentiated Services (DS) field of the Internet Protocol (IP) packet header in an OSI or TCP/IP model, which prioritizes different types of packets for forwarding to the router based on the attached payload. Similarly, DPI plays a crucial role in network management, enabling the identification of applications, services, and potential threats within the network traffic. In this study, various machine learning models, namely Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, Random Forest, Logistic Regression and ensemble models such as, XGBoost, AdaBoost were used to harness the capabilities of network packet classification based on DSCP. Detailed experimentation was conducted to evaluate their performance. The results show that AdaBoost demonstrated superior performance with an accuracy of around 89.91%, showcasing its ability to adapt the evolving network configurations and conditions while maintaining high classification accuracy on the IP packets. The random forest model also performed well, achieving an accuracy of 89.41%, making it a strong candidate for the DSCP classification in network transmission. This study has the potential to significantly improve how networks manage traffic, prioritize packets, and secure complex and dynamic network environments.

show abstract

DeepMatch

Cited by 32 publications

References 56 publications

Optimizing Application Performance with BlueField: Accelerating Large-Message Blocking and Nonblocking Collective Operations

Optimizing Application Performance with BlueField: Accelerating Large-Message Blocking and Nonblocking Collective Operations

P4DDPI: Securing P4-Programmable Data Plane Networks via DNS Deep Packet Inspection

Machine Learning-based Enhanced Deep Packet Inspection for IP Packet Priority Classification with Differentiated Services Code Point for Advance Network Management

Contact Info

Product

Resources

About