CloudStack Architecture Future
7 likes2,599 views
The document summarizes CloudStack architecture plans for the future. It discusses moving to management server clusters per availability zone rather than per region. It also discusses using an object storage system for templates and snapshots rather than a separate NFS server. Finally, it discusses a possible future model where CloudStack manages existing virtualization clusters rather than deploying and managing its own system VMs.
CloudStack Architecture Future
- 1. CloudStack Architecture Future Sheng Liang, CTO Cloud Platforms, Citrix Systems July 21, 2012
- 2. Users SS CP Cloud VR EC2 Stack API Layer VM VM Access control System VM Framework Orchestration Plug-in Engine Framework Event Framework Resource Management Plug-ins Message Bus (for storage, vSphere XenServer KVM SRX NetScaler Other networking, Resource Resource Resource Resource Resource Resources hypervisor, etc.)
- 3. Future: CloudStack storage CloudStack Primary Storage Mgmt Server • Can be configured zone-wide • Supports true EBS implementations AWS-style AWS-style AWS-style Availability Availability Availability Zone Zone Zone Secondary Storage • No longer needed zone-wide NFS server Object Storage • Region-wide object store for templates, ISOs and snapshots • Templates and ISOs cached in Primary Storage
- 4. • Apache Foundation project • Hadoop Core includes: ᵒ HDFS - Distributed file system derived from Google File System ᵒ Map/Reduce – Distributed compute engine • Written in Java • Runs on Linux, Mac OS/X, Windows, and Solaris • Optimized for commodity hardware
- 5. Now: management server clusters, zones, and regions Region 1 Region 2 Data Data Data Data Data Data Center Center Center Center Center Center Primary Standby Primary Standby Mgmt Mgmt Mgmt Mgmt Server Server Server DR Server Cluster DR Cluster Cluster Cluster Availability Availability Availability Availability Availability Availability Zone Zone Zone Zone Zone Zone Object Storage Object Storage
- 6. Future: management server cluster per zone Region 1 Region 2 Data Data Data Data Data Data Center Center Center Center Center Center Mgmt Mgmt Mgmt Mgmt Mgmt Mgmt Server Server Server Server Server Server Cluster Cluster Cluster Cluster Cluster Cluster Availability Availability Availability Availability Availability Availability Zone Zone Zone Zone Zone Zone Object Storage Object Storage
- 7. • Implement rigorous state machines for all operations • Customizable workflow engine • Enhance plug-in framework ᵒ Storage plug-in
- 8. • Stateless, recreatable, and restartable • CloudStack manages auto-scaling of System VMs • System VMs may be dedicated (e.g., virtual router) or shared (SSVM and CPVM) • Potentially expose System VM framework to 3rd-party VMs? ᵒ NetScaler VPX ᵒ User VM
- 9. S3 API NFS, CIFS, HDFS, 3rd-Party Object S3 API VM Store API, … S3 API VM Object Store … … … (e.g., NetApp, Hadoop, Cloudian) S3 API VM CloudStack Compute Nodes
- 10. • Clean up access control code • API servers run as a separate service ᵒ Running in System VM
- 11. • Built on message bus • Treats management server DB as the desired state • Some downsides: ᵒ Prevents use of native virt mgmt tools ᵒ Bugs could cause disastrous results (like stopping/destroying guest VMs) • An alternative model: CloudStack functions as a virt admin user, enabling super admin to use native virt mgmt tools
- 12. Future model Userh CloudStack Mgmt Server vCenter ESXi Cluster vCenter Client Admin XenServer Cluster XenCenter Hyper-V Systems Cluster Center
- 13. • Register existing virt clusters • CloudStack must assume any of the underlying networks, storage, hosts, and VMs can change state any time • Expose additional virt mgmt capabilities as cloud API
- 14. Manage existing server virt clusters CloudStack Region AWS-style AWS-style AWS-style Existing Existing Availability Availability Availability Server Virt Server Virt Zone Zone Zone Clusters Clusters Object Storage
- 15. Work better. Live better.