Machine Learning Based Digital Intelligence for MultiCloud

In Kubernetes (K8s), after Day 1 Deployment, we need to consider the requirements from Day 2. Machine Learning Workload prediction can help us to optimize the user experience, minimize the cost, and enhance the utilization.

We are working on Federator.ai to deliver a solution for:

Monitoring (for performance and issues)
Resource Orchestration (where to put the pots/clusters on nodes, considering their communication, workloads, and duration of services)
Automation with Optimization (continuous engagement with the deployment and autoscaling recommendation) -- Point Decision Making to Foresight Optimization
Performance Enhancement without code change
Just in Time Fitted based on ML-Digital Intelligence
Economic Arbitrage (Demand Side Platform)
Issue Isolation and Resolution (MultiLayer correlation for MultiCloud)
Fixes Validation

Transcript

1. Machine(Learning(Based(Digital( Intelligence)for)MultiCloud Eric%Chen,%CEO Eric.Chen@prophetstor.com Twitter:(ericchen1010 ProphetStor)Data)Services,)Inc. 1
2. Who$Am$I? 2
3. Cloud&Native&Fuels&MultiCloud 3
4. IBM$Acquired$Red$Hat$for$ OpenShift/MultiCloud 4
5. K8S$Day$1$and$Day$2 5
6. Day$1 Deployment:+Packaging+and+Automation Containerized+Applications Communications+and+Security+among+pods/nodes/clusters Private(or(Public(Clouds? 6
7. Day$2 • Monitoring((for(performance(and(issues) • Resource(Orchestration((where(to(put(the(pods/clusters)on)nodes,) considering*their*communication,*workloads,*and*duration*of*services) • Automation)with)Optimization)(continuous)engagement)with)the) deployment*and*autoscaling*recommendation)*!! Point&Decision&Making&to& Foresight*Optimization • Performance*Enhancement*without*code*change • Just%in%Time%Fitted%based%on%ML!Digital'Intelligence • Economic'Arbitrage'(Demand'Side'Platform) • Issue%Isolation%and%Resolution%(MultiLayer correlation*for*MultiCloud) • Fixes&Validation& 7
9. 9 When%does%ProphetStor%add%value%in%Kubernetes? Build&and&deploy& applications SLA$and$cost$ analysis Deploy' Kubernetes Operation Container)scaling) configuration Scheduling Initial'deployment' planning Self!learning(pod( scaling(configurations Policy!driven' predictive)pod) placement Cross!layer&anomaly& detection,)hardware) health&detection Kubernetes) autoscaler IT#administrator VPA CA HPA Assist Assist` Day!1"Deployment Day!2"Operations Cost!,"locality!," and$SLA! aware% node%predictive% scaling Assist Assist Application!aware NetApp& Kubernetes)Service Federator.ai*improves*Kubernetes*operational*efficiency*for*day!2"operations," optimizes)resource)utilization)and)saves)costs.)
10. Architecture)(Using)OpenShift)As)An)Example) OS Virtualization Containers OpenShift CPU Memory GPU Network Storage Private(Cloud Hybrid'Cloud Public'Cloud Workload(Prediction Anomaly(detection Cross!layer&correlation Cloud&Arbitrage Cloud&instance& recommendation Resource( Planning Cost%Optimization% Dashboard API Cloud&Portal Alerting Federator.ai Kubernetes) Platform 10
11. Example:)Working)with) Red$Hat$OpenShift 11
13. Deployment*in*the*Cluster OpenStack Hardware'(CPU) Hardware'(GPU) VMs OpenShift*(OKD) Federator.ai AIOps HPC$Cluster GPFS Ceph Kubernetes Prometheus InfluxDB GPU$Workload 13
15. Example:)Working)with) NetApp&Kubernetes&Services 15
16. ©"2019"by"ProphetStor"Data"Services,"Inc. 16 Federator.ai*in*NetApp*Gallery Install'Federator.ai'from' “Solutions”+!>"“Trusted" Charts”
18. ©"2019"by"ProphetStor"Data"Services,"Inc. 18 Cost%Saving%of%Day!1"Deployment AWS Current: Asia Pacific (Tokyo) - master: t2.large (2 vCPUs, 8GB) x 1 - worker: c4.8xlarge (36 vCPUs, 60GB) x 2 Recommendation: US West (Oregon) - master: t2.large (2 vCPUs, 8GB) x 1 - worker: c5.9xlarge (36 vCPUs, 72GB) x 2 24%$Saving
19. ©"2019"by"ProphetStor"Data"Services,"Inc. 19 Cost%Saving%of%Day!1"Deployment AWS Current: Asia Pacific (Tokyo) - master: t2.large (2 vCPUs, 8GB) x 1 - worker: c4.8xlarge (36 vCPUs, 60GB) x 2 GCP Recommendation: US West 1c - master: n1-high-mem-2 (2 vCPUs, 13GB) x 1 - worker: n1-standard-8 (8 vCPUs, 30GB) x 9 40%$Saving
20. Resource/Cost+Monitoring+and+ Prediction*of*Day!2"Operations
21. 21 Resource/Cost+Monitoring+and+Prediction+of+Day!2"Operations Resource(Utilization/ RI Recommendation/ Cost%Analysis%Charts
22. ©"2019"by"ProphetStor"Data"Services,"Inc. 22 Historical*&*Predicted*Resource*Utilization Historical/Predicted.resource.utilization.and.resource.idleness
23. ©"2019"by"ProphetStor"Data"Services,"Inc. 23 Recommendation+for+Reserved+Instance+Commitments RI#recommendation#based#on#predictions#of#current#provider# and$configuration (same&instance&type,&number&of&nodes,& region,(and(etc.)
24. ©"2019"by"ProphetStor"Data"Services,"Inc. 24 Cost%Analysis%for%Cloud%Services Cost%analysis%for%different%providers%and%configurations%with%RI%optimization
25. Per$container$resource$recommendations 25
26. 26 Per$container$resource$recommendations
27. Per$application$resource$recommendations 27
28. Per$namespace$resource$recommendations 28
29. Cluster!wide%resource%recommendations 29
30. 30 Cluster(Health(Monitoring
31. 31 Node%health%monitoring
32. 32 Resource(Prediction
33. 33 Federator.ai*HPA*reduces*Nginx*latency
34. 34 Federator.ai*HPA*Reduces*PostgreSQL*latency
36. We#are#hiring! 36
37. Predict(GPU(workload(based(on(historical( GPU$workload$metrics 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 0:00 GPU Duty Cycle (%) GPU$Container$ workload( prediction Just!fitted&GPU&resource& recommendation+for+the+ container 37
38. Recommend(container(GPU(resources(and( pod$placement$based$on$prediction 38 GPU Resource Recommendation Container A Container B Container C