OpenStack中通过pci passthrough集成GPU

1. 控制节点中的nova配置文件

[DEFAULT]
pci_alias = {"name":"K40m_Tesla","vendor_id":"10de","product_id":"1023"}
scheduler_available_filters = nova.scheduler.filters.all_filters 
scheduler_available_filters = nova.scheduler.filters.pci_passthrough_filter.PciPassthroughFilter 
scheduler_default_filters = RetryFilter,AvailabilityZoneFilter,RamFilter,ComputeFilter,ComputeCapabilitiesFilter,ImagePropertiesFilter,ServerGroupAntiAffinityFilter,ServerGroupAffinityFilter,PciPassthroughFilter
scheduler_driver = nova.scheduler.filter_scheduler.FilterScheduler
scheduler_max_attempts = 5

2. 计算节点中的nova配置文件

[DEFAULT]
pci_passthrough_whitelist = {"vendor_id":"10de","product_id":"1023"}

3. 创建GPU实例模板

openstack flavor create --public --ram 2048 --disk 40 --vcpus 2 m1.large.2xK40m
openstack flavor set m1.large.2xK40m --property pci_passthrough:alias='K40m_Tesla:2'

4. 安装K40m驱动

在虚拟机实例中安装驱动，驱动下载地址:点击打开链接。

5. 测试GPU

通过执行nvidia-smi命令查看GPU使用情况。