以下列表提供了有关最重要的 vSphere Bitfusion 命令及其任务的信息。如有必要,VMware 技术支持团队还会提供其他 CLI 命令。

vSphere Bitfusion 中分配 GPU

要为单个应用程序分配多个 GPU,请运行 bitfusion run 命令。

要分配多个 GPU 并启动会话以便在相同 GPU 上运行多个应用程序,请运行 bitfusion request_gpus

vSphere Bitfusion 中运行应用程序

要启动单个应用程序,请运行 bitfusion run 命令。

要在使用 bitfusion request_gpus 命令启动的会话中启动多个应用程序,请运行 bitfusion client 命令。

vSphere Bitfusion 中解除分配 GPU

要在使用 bitfusion request_gpus 命令启动的会话中解除分配 GPU,请运行 bitfusion release_gpus 命令。

列出 vSphere Bitfusion 中的可用 GPU

要验证 vSphere Bitfusion 服务器安装并查找可用 GPU 列表,请运行 bitfusion list_gpus 命令。

 - server 0 [172.16.31.162:56001]: running 0 tasks
   |- GPU [0]: free memory (15109 / 15109MiB) Tesla T4 (7.5)
 - server 1 (leader)  [172.16.31.156:56001]: running 0 tasks
   |- GPU [0]: free memory (15109 / 15109MiB) Tesla T4 (7.5)

vSphere Bitfusion 中执行运行状况检查

可以从命令行访问运行状况检查。
  • 要检查所有 vSphere Bitfusion 服务器和 Bitfusion 客户端的运行状况,请运行 bitfusion health
  • 要检查单个 vSphere Bitfusion 客户端或服务器的运行状况,请运行 bitfusion localhealth

检查 vSphere Bitfusion 版本

要显示 vSphere Bitfusion 的安装版本,请运行 bitfusion version 命令。

Bitfusion version: 4.0.0 release

vSphere Bitfusion 中显示 GPU 信息

要显示 GPU 信息,请运行 bitfusion smi 命令。或者,要收到类似的输出,可以使用 bitfusion run 命令启动 nvidia-smi 应用程序。

+----------------------------------------------------------------------------------------+
| 172.16.31.162:56001                                          Driver Version: 460.73.01 |
+--------------------------------------+-------------------------+-----------------------+
| GPU  Name              Persistence-M | Virt Mem    Alloc / All | BusId  Vol Uncorr ECC |
| Fan  Temp  Perf        Pwr:Usage/Cap | Phy Mem     Used  / All | GPU-Util   Compute M. |
|======================================+=========================+=======================|
| 0    Tesla T4               Enabled  | 0       MB / 15109   MB | 00000000:13:00.0    0 |
| 0 %   28C  P8             10W /  70W | 3       MB / 15109   MB |   0%          Default |
+--------------------------------------+-------------------------+-----------------------+
+----------------------------------------------------------------------------------------+
| 172.16.31.156:56001                                          Driver Version: 460.73.01 |
+--------------------------------------+-------------------------+-----------------------+
| GPU  Name              Persistence-M | Virt Mem    Alloc / All | BusId  Vol Uncorr ECC |
| Fan  Temp  Perf        Pwr:Usage/Cap | Phy Mem     Used  / All | GPU-Util   Compute M. |
|======================================+=========================+=======================|
| 0    Tesla T4               Enabled  | 0       MB / 15109   MB | 00000000:13:00.0    0 |
| 0 %   34C  P8             10W /  70W | 3       MB / 15109   MB |   0%          Default |
+--------------------------------------+-------------------------+-----------------------+

vSphere Bitfusion 中测试带宽

要测试 vSphere Bitfusion 客户端和服务器之间的带宽和延迟,请运行 bitfusion net_perf 命令。

单个网络接口
Displayed results are calculated from round-trip measurements
BW(1MB) = 1000/(LAT(1MB) - LAT(1B))

[ <client>] ens160 => [10.202.8.169] net1 ( tcp) Single packet lat = 51 us, bw(1MB) = 1.71 GB/s
[ <client>] ens160 => [10.202.8.185] net1 ( tcp) Single packet lat = 48 us, bw(1MB) = 1.09 GB/s
[ <client>] ens160 => [10.202.8.233] net1 ( tcp) Single packet lat = 50 us, bw(1MB) = 0.87 GB/s
多个网络接口
Displayed results are calculated from round-trip measurements
BW(1MB) = 1000/(LAT(1MB) - LAT(1B))

[ <client>] ens160 => [10.202.8.169] net1 ( tcp) Single packet lat = 51 us, bw(1MB) = 1.71 GB/s
[ <client>] ens160 => [10.202.8.185] net1 ( tcp) Single packet lat = 48 us, bw(1MB) = 1.09 GB/s
[ <client>] ens160 => [10.202.8.233] net1 ( tcp) Single packet lat = 50 us, bw(1MB) = 0.87 GB/s
[ <client>] ens192f0 => [10.202.8.169] net2 ( tcp) Single packet lat = 47 us, bw(1MB) = 2.14 GB/s
[ <client>] ens192f0 => [10.202.8.185] net2 ( tcp) Single packet lat = 49 us, bw(1MB) = 1.11 GB/s
[ <client>] ens192f0 => [10.202.8.233] net2 ( tcp) Single packet lat = 50 us, bw(1MB) = 1.15 GB/s
[ <client>] vmw_pvrdma0 => [10.202.8.169] vmw_pvrdma0 (infiniband) Single packet lat = 19 us, bw(1MB) = 3.66 GB/s Single packet Write lat = 8 us, bw = 10.101 GB/s
[ <client>] vmw_pvrdma0 => [10.202.8.185] vmw_pvrdma0 (infiniband) Single packet lat = 21 us, bw(1MB) = 3.45 GB/s Single packet Write lat = 8 us, bw = 10.5263 GB/s
[ <client>] vmw_pvrdma0 => [10.202.8.233] vmw_pvrdma0 (infiniband) Single packet lat = 21 us, bw(1MB) = 3.46 GB/s Single packet Write lat = 8 us, bw = 10.4167 GB/s

vSphere Bitfusion 中请求帮助

要获取 vSphere Bitfusion CLI 命令的完整列表或有关特定命令的详细信息,请运行 bitfusion help 命令。

NAME:
   Bitfusion - Run application with VMware Bitfusion

USAGE:
   bitfusion <command> <options> "application"
   bitfusion <command> <options> -- [application]
   bitfusion help [command]

   For more information, system requirements, and advanced usage please visit docs.bitfusion.io

COMMANDS:
        tls-certs, TC    Manage TLS certificates used by bitfusion server.  Requires root privileges.
        version, v       Display full Bitfusion version
        localhealth, LH  Run health check on current node only
        dealloc          Deallocate license certificate.  Requires root priviledges.
        crashreport      Send crash report to bitfusion
        list_gpus        List the available GPUs in a shared pool
        initdb           Init database setup
        token            Fetch and manipulate tokens
        register         Register remote server as the plugin
        unregister       Unregister remote plugin
        removenode       Remove unavailable nodes
        user             Manage bitfusion users
        help, h          Shows a list of commands or help for one command
   Client Commands:
        client, c     Run application
        health, H     Run health check on all specified servers and current node
        request_gpus  Request GPUs from a shared pool
        release_gpus  Release GPUs back into a shared pool. Options must match a previous request_gpus command
        run           Request GPUs from a shared pool, run a client command, then release the GPUs
        stats         Gather stats from all servers.
        smi           Display smi-like info for all servers.
        local         Run a CUDA application locally
        net_perf      Gather network performance data from all SRS servers.
   Server Commands:
        server, s                Run dispatcher service - listens for 'bitfusion client' commands
        resource_scheduler, srs  Run Bitfusion resource scheduler (SRS) on GPU server
        analytics                Run Bitfusion analytics server
        manager                  Run Bitfusion manager server

EXAMPLES:
   $ bitfusion resource_scheduler --srs_port 50001

   $ bitfusion run -n 4 -- <application>

   $ bitfusion request_gpus -n 1 -p 0.25