Gorillas in the mist
•Download as PPTX, PDF•
1 like•602 views
OpenStack Summit, Vancouver, 2015 talk Title: Gorillas in the mist - Enterprise Workloads on OpenStack
Report
Share
Gorillas in the mist
- 1. Gorillas in the mist Enterprise Workloads on OpenStack
- 2. Gerd Prüßmann Cloud Architect Deutsche Telekom AG @2digitsLeft g.pruessmann@telekom.de Sriram Subramanian Founder & Principal Cloud Specialist CLOUDDON @sriramhere sriram@clouddon.com
- 3. Agenda Brief History Toolbox Lessons Learnt Deutsche Telekom Overview Overview of Workloads Workload #1 Workload #2 Lessons Learnt Next Steps
- 4. Brief History ● Compute Intensive ○ E Commerce ○ Research ○ Service Provider ○ Streaming ● Large Storage ○ Media ● Compute + Storage ○ Bio Informatics ○ Analytics
- 5. Toolbox ● Cloud Native Applications Unicorns ● Port ● Rewrite/ Partial Rewrite ● Smart Packaging ● Backoff :)
- 6. Lessons Learnt ● Start from Workloads ● No One Rule Fits All ● Wrapping in VM != Cloud ● It’s OK to leave some out
- 7. • Business Market Place (BMP) • https://portal.telekomcloud.com/ • SaaS offering - applications from Software Partners (ISVs) and DT offered to SME end customers • Cloud platform based on Open Source technologies only (OpenStack, CEPH, Ubuntu) • Project started early 2012 with OpenStack Essex, CEPH • 1st OpenStack platform in production since Q1/2013 • complements other platforms: Enterprise clouds based on i.e. VMware / SAP Hana, OpenStack based Cisco “Intercloud”, OS platforms for NFV, research, IaaS Deutsche Telekom Overview
- 8. OVERVIEW OF WORKLOADS • Enterprise Social Network • Enterprise Cloud Storage TeamDisk • Payroll Management Sage • Enterprise Resource Planning SilvERP • Customer Relationship Management • Invoice Management Fastbill • Knowledge Management • Enterprise Content Management Lexmark Enterprise Software, EasySyS • Project Management Projecterus, Teamlike, weclapp • Contract Management • Document Management Mobile Devices PadCloud
- 9. all applications are completely different w.r.t. • technologies used • tenant size / resource usage • scalability / elasticity • business case • number of users / target groups • operational maintenance efforts • level of cloud awareness OVERVIEW OF WORKLOADS
- 10. SERVICE MODELS • Two models: Managed model vs. Hosting model • Cloud resources OS Tenant, instances, storage (RAM, persistent Volumes / S3) • Reference and production tenants • PaaS services / support VPN, load balancer, proxy server, email gateways, diff. databases, mirror, puppet master, DNS, NTP, backup, monitoring • Individual integration / onboard support
- 11. WORKLOAD - BUSINESS CLOUD STORAGE • Application developed by DT • Enterprise secure online storage for documents, images etc. • web app, mobile app, PC SyncClient available • Petabyte storage cluster (CEPH) • integrated in some other applications on BMP (easily exchange documents between apps) • bundled with every business users mobile phone contract
- 12. WORKLOAD - BUSINESS CLOUD STORAGE Application consists of • Apache web servers (memcached) • JAVA application servers • ActiveMQ servers • Transcoders (image processing) • MongoDB servers • Load balancers
- 13. WORKLOAD - BUSINESS CLOUD STORAGE Before onboarding: Non-dynamically scalable installation on physical servers • Traditional 3-tier deployment web-, application-, database servers (master-slave), • DRBD replication • Deployed on multiple huge servers in DC • Expensive storage appliances (NFS) • no configuration management / automated installation
- 14. WORKLOAD - BUSINESS CLOUD STORAGE After onboarding: dynamically scalable tenant on OpenStack • Load balancers in front of each layer (web service / application server / API) • Deployed on multiple standard KVM VMs on OpenStack • Multiple MongoDB database servers • all persistent volumes on CEPH RBD • NFS replaced by object storage (CEPH S3) • configuration managed by Puppet
- 15. WORKLOAD - BUSINESS CLOUD STORAGE TOOLBOX • small partial rewrite (cloud S3 storage backend, MongoDB) • smart deployment (Load balancers) • configuration management (puppetized) / automatic installation introduced • scalable and highly available storage backend introduced (CEPH, RBD/S3) Result: Scalable enterprise application on cloud
- 16. WORKLOAD - ENTERPRISE SOCIAL NETWORK • ISV’s enterprise social network application for private cloud offering • originally “on premise” solution • huge customer specific modifications • multiple Single-Instances for private cloud customers (just one VM for every App Tenant) • “tenant VM” contains application and services • every new “customer tenant” gets a single, new VM
- 17. WORKLOAD - ENTERPRISE SOCIAL NETWORK onboarding actions: • ISV implemented / replaced storage backend with S3 (CEPH) • configuration management (puppetized) introduced • “cloud manager” service introduced by the ISV to provision, reboot VMs (single “customer tenants”) First user login starts the instance - last logout closes down VM • “cloud manager” portable to any cloud platform!
- 18. WORKLOAD - ENTERPRISE SOCIAL NETWORK TOOLBOX • partial rewrite - adoption of cloud S3 storage backend • additional development - cloud manager • smart deployment with “containerized” VMs to prevent heavy changes on a pet / on premise app • accepted limitations of scalability and availability (VM reboots <1min)
- 19. LESSONS LEARNT • Enterprise apps: no cattles - only pets • Enterprise apps / sector software: specific target groups / limited number of users / not millions of users worldwide • “Cloud native” from ISVs point of view: rewrite app - huge investments - no business case - no ROI • Majority of enterprise apps are not well prepared for cloud (Legacy, stateful, not cloud aware, no DevOps technologies)
- 20. LESSONS LEARNT • Initially increased integration efforts on cloud or OpenStack • Good toolset to ease the migration to cloud: Partial app rewrite (to integrate cloud technologies) changed deployment and distribution architecture configuration management change processes • Adjust installation, maintenance processes of the apps
- 21. Next Steps ● App Ecosystem Work Group ● Win the Enterprise Workgroup ● OpenStack User Stories - SuperUser ● Cloud Native Applications Whitepaper
- 22. Gerd Prüßmann Cloud Architect Deutsche Telekom AG @2digitsLeft g.pruessmann@telekom.de Sriram Subramanian Founder & Principal Cloud Specialist CLOUDDON @sriramhere sriram@clouddon.com
Editor's Notes
- Introduce Speakers Talk about their experience and expertise
- Walk over the agenda. Sriram will provider a brief history of variety of workloads people have had success running on OpenStack. It includes a variety of applications, not just enterprise specific Sriram will talk about common tools used to develop or migrate workloads to cloud, following up with lessons learnt. Gerd will then talk about their private cloud and some of the enterprise workloads DT have had success with. For each workload, Gerd will provide details on what the aplication is, what is the environment, how they moved to cloud and any lessons learnt. Finally, we will summarize the discussions, followed by providing next step and a call for action to join the Application Development Work Group.
- Starting from early days of OpenStack when we only had compute, storage and network services, people have been running variety of workloads. This was also the time when most important use case of OpenStack was cloud service providers. Remember we are talking about the days when it was still largely nova-network and quantum. Neutron mess was yet to be created and cleared! Early use cases where either compute intensive, spinning up lots of VMs or large storage. Compute Intensive workloads suchs ecommerce applications, web services, high performance research fall under this category. You must’ve heard about WalMart and PayPal during the keynotes. PayPal/ eBay were one of the early adopters. They were running their services powering the Shopping Cart experience including credit, payments and merchants operations. WalMart had all of their WalMart.com portal on OpenStack cloud which consists of variety of web services. CERN is a great example for HPC use case. They are also one of the largest OpenStack clouds A relatively mature Object store Swift enabled a lot of workloads that demanded cheap, large storage. Workloads included Media storage, repositories, catalogs for ecommerce vendors. MercadaLibre, largest ecommerce portal in Brazil were a good example. Around the same time, there was increasing interest from TelCo service providers and Neutron service started to stabilize.
- If you’re lucky, all your applications are cloud native or cloud aware. Not often
- Introduce Speakers Talk about their experience and expertise