Release v23.4

HAProxy is an open-source software load balancer and reliable reverse proxy server. This workload is combined with Intel® Quick Assist Technology to speed up the performance for web security access.

Nginx DLB workload is used to measure the latency (in milliseconds) and RPS (requests per second) performance of Nginx w/ and w/o Intel® Dynamic Load Balancer. To improve the performance, NGINX introduces AIO thread module and in this workload, Intel® DLB hardware queue is used to replace the software queue in the AIO thread module.

Calico VPP workload is a Kubernetes CNI for benchmarking the throughput performance benefits of DSA for Calico VPP. DSA accelerates the memory copy operations for Calico VPP memif interface, thus improve throughput with less cores.

Istio Envoy forward Proxy workload is a crucial component in the Istio service mesh architecture that plays a pivotal role in handling outbound traffic from microservices to external destinations. As a type of proxy, it acts as an intermediary between microservices and external services or APIs, facilitating communication while providing additional features for security, observability, and control.

Edge-Ceph-VirtIO workload is a cloud native workload for benchmarking the virtualized IO on cloud native ceph (rook-ceph). It is developed based on Kubevirt, Ceph and SPDK, it enables and optimizes high performance virtualization IO for cloud native Ceph storage on edge platform, support VM live-migration and live-recovery.

FFMPEG x264 workload is a media transcode workload with FFmpeg and x264. FFmpeg is the leading multimedia framework, able to decode, encode, and transcode video files. The workload supports FFmpeg version 4.4 and 6.0, with x264 preset: fast, medium, very slow. To measure the transcoding capability of the system, it launches multiple FFmpeg instances to fully utilize the CPU resources.

Spark is a decision support benchmark that models several generally applicable aspects of a decision support system, including queries and data maintenance. The benchmark provides a representative evaluation of performance as a general-purpose decision support system. A benchmark result measures query response time in single user mode, query throughput in multiuser mode and data maintenance performance for a given hardware, operating system, and data processing system configuration under a controlled, complex, multi-user decision support workload.

Cassandra workload assesses the performance of Apache Cassandra using Apache Cassandra built-in application (cassandra-stress). In WSF, operations per second metric is measured to gauge the performance of Apache Cassandra.

MongoDB workload is an evaluation of MongoDB's performance using the YCSB framework, a widely used benchmarking tool. MongoDB is a NoSQL database system that is known for its flexibility, scalability, and ease of use. The YCSB framework measures the performance of database systems under various workloads, simulating real-world scenarios.

Kafka workload assesses performance using built-in application within Apache Kafka. Apache Kafka is a framework implementation of a software bus using stream-processing and the test case in WSF measures its producer performance.

Malconv workload is a deep learning model designed for malware detection and classification which uses a convolutional neural network to analyze the raw binary data of a file. The workload supports execution on TensorFlow and ONNX.

Istio Envoy is the workload leverages Istio with Envoy as sidecar, which is the industry widely adopted Service Mesh implementation. With sidecar introduced as the data plane networking proxy, the achieved RPS and latency performance was the interest of customer in internet service providers. The workload benchmark shows the performance of service mesh with tuned Istio Envoy Gateway also with/without Intel accelerators (QAT, CryptoMB, DLB...) on IA platforms.

Release v23.3

• The Video Structure workload is an AI-based solution for object detection and object classification, using YOLO model as the object detector and ResNet50 model as the object classifier.
• The CDN -NGINX workload is a network of servers linked together with the goal of delivering content as quickly, cheaply, reliably, and securely as possible.

The Video Structure and CDN-NGINX workloads are validated on top of Network and Edge Platform (NEP) Reference Systems . For best performance, the On_Prem_VSS profile shall be used for Video Structure and On_Prem for CDN.

• The SPDK- NVMe-o-TCP workload builds nvme-over-tcp storage target with SPDK and leverages Linux kernel nvme/tcp as an initiator for IO benchmarking.
• FIO stands for Flexible IO. It is an IO benchmark workload that simulates a given IO workload on a specific platform.
• Istio Envoy is a workload that leverages Istio with Envoy as sidecar, which is the industry widely adopted Service Mesh implementation.
• The Linpack workload contains Intel’s distribution for the High-Performance Computing LINPACK Benchmark, which is a library for performing numerical linear algebra to measure a system's floating-point computing power.
• The SpecCpu-2017 workload contains CPU intensive suites for measuring and comparing compute intensive performance, stressing a system’s processor, memory subsystem and compiler.
• The 3DHuman-Pose-Estimation workload is designed to estimate 3D human pose from a single RGB image, which consists of human body detector based on YoloX, 2d human pose estimator based on HRNet, and 3d human pose estimator based on HMR.
• Intel® Deep Link Workload Scheduler is designed as a toolkit, which can do workload pipeline profiling, runtime telemetry monitoring and the pipelines management/ scheduling on heterogeneous computing platform.

Release v23.2

The v23.2 release includes the following features/updates:

• Validated NGINX, OpenSSL3, and Malconv with Network and Edge Platform Reference Architecture. For best performance, use the Remote Central Office-Forwarding profile for NGINX and OpenSSL3 and Basic for Malconv.
• The Edge-Ceph-VirtIO workload implements block storage functions using the Ceph storage backend. The workload is accelerated by the Intel® SPDK. The workload measures the libvirt virtualized I/O performance, one of the key performance indicators for Edge-Native applications.
• The CM-xAPP-OpenVINO workload is an Intelligent Connection Management (CM) xApp developed on the O-RAN network architecture using the Intel® Distribution of OpenVINO™ toolkit to optimize user association and load balancing. The workload improves the quality of service (QoS) of an User Equipment (UE). This workload is under NDA. Please reach out to wsf-externalrelease-communicatio for further support.
• The OpenSSL3 workload measures the OpenSSL* software performance. The workload is integrated with the Intel QuickAssist Technology software library.
• The NGINX workload was updated to utilize the Intel QuickAssist Technology.

Release 23.1.1

This is the minor release of 23.1 with the following features:

• New software stack: PyTorch Xeon
• New workload: BertLarge-Pytorch-Xeon-Public, Resnet-50-Pytorch-Xeon-Public.
• Fix bugs

Release 23.1

This WSF release includes the following new workloads (and bug fixes of existing workloads):

• Smart science lab:
  • The workload can be used to quickly and accurately grade students' performance in a physics experiment. The workload uses FFmpeg (with VA GPU acceleration) to decode video, temporal segmentation algorithms to locate key frames, and advanced object detection algorithms to identify key objects.

Some features are available only under NDA.

• SpecJBB:

  • The workload is the SpecJBB* 2015 benchmark, which measures performance of the latest Java* application features. Support for OpenJDK* and Zulu* JVMs.

• Stream:

  • The Stream benchmark is a simple synthetic benchmark program that measures sustainable memory bandwidth (in MB/s) and the corresponding computation rate for simple vector kernels.

• Mongo-YCSB:

  • The workload uses the Yahoo! Cloud Serving Benchmark (YCSB) to measure the MongoDB performance.

v22.4 Release

• New workloads: NGINX, MalConv, and iPerf.
• QAT crypto base image: Enable speed test of the OpenSSL software with QAT.
• The infrastructure code changed from cumulus to terraform.

Release 22.40.1

This is the first external release for Intel® Workload Services Framework Repository. The WSF repository contains a set of workloads that can be used to exercise multiple platforms. Each workload is a complete and standalone implementation that can be built and run collectively or individually. 3 workloads are onboarded: HammerDB MySQL, Kafka and OpenSSL QAT.