Guaranteeing Performance in High- Density OpenStack Environments Endre Sara – VP of Dev @VMTurbo 1
OpenStack Workload Management Challenges • Limited workload scheduler • Lack of real-time control • VM placement decisions are static and limited in intelligence • Application and VM performance is not assured • Enabling OpenStack with KVM as a viable Enterprise Private Cloud • VMTurbo enhances OpenStack with intelligent workload management and control 2
VMTurbo’s Control Platform Guaranteeing Application Performance DEMAND SUPPLY
VMTurbo Use-Cases • IaaS lifecycle • Intelligent place and control workloads by integrating VMTurbo with the native OpenStack scheduler • Tiered service offerings • Guaranteeing performance of select workloads during periods of peak utilization while increasing overall efficiency of resources • Software license cost minimization • Limiting workload placement to specific sockets to minimize socket- based license costs 4
VMTurbo Use-Cases • VM groups - end-user identified inter-dependencies: • Affinity rules and Anti-affinity rules - enabling instance redundancy for High Availability or load balancing • "Shared nothing" live migration • Enabling use of cheaper local storage for ephemeral backed VMs • Preserving the ability to doVM live migration for Host maintenance, performance or efficiency • Host maintenance mode • Evacuating all instances from a Host to enable operational maintenance of Compute nodes 5
OpenStack with VMTurbo VMTurbo VMTurbo VMTurbo Nova Cinder usage Scheduler Scheduler metrics VMTurbo Operations Manager 6
VMTurbo Operations Manager 7
VMTurbo Contribution • Modify OpenStack (code, APIs, SDK) to discover, monitor and control OpenStack infrastructure and workloads on par with VMware workloads • Use the new OpenStack APIs in VMTurbo Operations Manager • Contribute the code to open source, so that customers may. . . • Benefit for better OpenStack visibility and management • Easily install VMTurbo enhancements to various versions of OpenStack easily installing diffs from our GitHub repository 8
VMTurbo OpenStack Contribution PROJECT FEATURE DESCRIPTION LINK Ceilometer Added Disk Meters for Libvirt Storage meters per instance Blueprint disk.capacity = Disk Capacity for all the disks of the instance Review 145819 Ø disk.usage = Disk Usage for all the disks of the instance Ø Specification disk.allocation = Disk Allocation for all the disks of the instance Ø Review Storage Meters per disk per instance disk.device.capacity = Disk Capacity for each individual disk of an instance Ø disk.device.usage = Disk Usage for each individual disk of an instance Ø disk.device.allocation =Disk Allocation for each individual disk of an instance Ø Ceilometer Added VM memory metric The 'memory.resident' metric will expose the physical hosts view of the amount of host Blueprint memory that is actually used to run the VM process Review 160866 Specification Review Documentation Ceilometer PM Memory usage metric (Bug Change SNMP 'hardware.memory.used' metricto reflect (memTotalReal - Review 157275 fix) memAvailReal) i.e. used value for the memory for a physical machine. Woorea Deserialize an empty json string Return null Image object in server description when server is not booted from Image GitHub pull 161 to a null object Woorea Added Java SDK for Added extension to provide pool information GitHub pull 184 scheduler_stats api in Cinder Pool names only: GET http://CINDER_API_ENDPOINT/v2/TENANT_ID/scheduler- Ø stats/get_pools Detailed Pools info: GET http://CINDER_API_ENDPOINT/v2/TENANT_ID/scheduler- Ø stats/get_pools\?detail\=True Woorea Add Hypervisor and Service class Added to Nova support: GitHub pull 188 support for Nova Ø Hypervisor, Hypervisors, HypervisorsResource Ø Service, Services, ServicesResource 9
Recommend
More recommend
