1、Harnessing the Power of AI/ML to Enhance SAI TestingTaras Chornyi,Director of Open Networking Solutions,PLVisionLeonid Khedyk,CTO,PLVisionHarnessing the Power of AI/ML to Enhance SAI TestingARTIFICIAL INTELLIGENCE(AI)AISAI defines a vendor-independent way of controlling forwarding elements.Common de
2、nominator-API for different types of networking ASICs:NPUPHYDPUExtensively used in NOSes:SONiCFBOSSDentOSproprietaryIntroductionCompliance and Standards a comprehensive and self-explanatory model of SAI is missingCoverage and Scalability available test content fails to identify functional and scalab
3、ility issuesDepth of testing inefficiencies in detecting complex networking issues(packet loss,latency,etc.)Results interpretation-demands in-depth knowledge of networking and comprehensive expertise in testing methodologies.Existing SAI testing solutions lack standardization,relevance,coverage and
4、usability.SAI Testing ChallengesIntroducing theconcept of AI-poweredtesting andits application inevaluating SAI implementations withtraffictests.Key test areas to focus:Network Topology ConfigurationGenerating Test CasesSAI ConfigurationTraffic PatternsScenario ExecutionTest Results AnalysisAI/ML-En
5、hanced SAI Testing ApproachTest definition as inputfor LLMGenerated test case injected into a frameworkSAI Challenger-wrapper framework to executeTest results/logs fed-back for interpretationBridging the Gap:AI/ML and SAI TestingDevices:DUT(Device Under Test):Type:RouterInterfaces:eth0(LAN),eth1(LAN
6、)IP Addresses:eth0-192.168.1.1/24,eth1-192.168.2.1/24MAC Addresses:eth0-00:11:22:33:44:55,eth1 00:11:22:33:44:56Host1:Interfaces:port1(connected to DUT eth1),port2(connected to Host1)IP Addresses:port1-192.168.1.2/24,port2-192.168.2.2/24MAC Addresses:port1-00:11:22:33:AA:55,port2