After you start a vSphere Bitfusion server instance, vSphere Bitfusion creates and maintains servers.conf and bitfusion-limits.conf configuration files on the client virtual machines (VMs). The client VMs must be deployed on ESXi hosts that are part of the same vCenter Server environment as the vSphere Bitfusion server instance.

Servers Configuration File

vSphere Bitfusion creates a high-priority user-specific file in ~/.bitfusion/servers.conf. Alternatively, you can create a system file /etc/bitfusion/servers.conf, which vSphere Bitfusion uses with a lower priority than the user-specific file. You use the cat command to display a server list.

To understand the command use, see the following example.

cat ~/.bitfusion/servers.conf

The servers configuration file lists the IPv4 addresses of all vSphere Bitfusion servers and ports that a vSphere Bitfusion client can access. The default port 56001 is not listed.
172.31.51.20
172.31.51.26:56003
172.31.51.42 56003
You can run the run command with an alternative vSphere Bitfusion server list that is a subset of the primary server list of GPU servers maintained by vSphere Bitfusion in the ~/.bitfusion/servers.conf file. To create a subset list of vSphere Bitfusion servers, you can perform one of the following steps. vSphere Bitfusion supports IPv4 addresses only.
  • You can use --servers value, -s value and supply a subset of the primary server list in a file of your choice. You must change the value argument with a filepath to a servers.conf file.
  • You can use --server_list value, -l value and supply a subset of the primary list of servers in the command line. You must change the value argument to a "ip_address:port;ip_address:port" format.

    You must enclose the list within quotes, because a semicolon is used as a separator when you list multiple addresses and the command-line interpreter can parse the list as multiple commands.

Limits Configuration File

The following limits apply to members of the vSphere Bitfusion group. Any user of the vSphere Bitfusion client must be a member of the vSphere Bitfusion group.

The bitfusion-limits.conf configuration file is installed on the vSphere Bitfusion client in /etc/security/limits.d/bitfusion-limits.conf by the client package. The file contains the following settings, which you can view and enforce by using the standard Linux utility, ulimit.
  • Maximum number of open files
    @bitfusion soft nofile 100000
    @bitfusion hard nofile 100000
  • Unlimited locked-in-memory address space
    @bitfusion soft memlock unlimited
    @bitfusion hard memlock unlimited
  • Unlimited maximum resident set size
    @bitfusion soft rss unlimited
    @bitfusion hard rss unlimited
Note:

If the resource limit for open files is too low, vSphere Bitfusion might receive a connection error: Cannot allocate memory error. To resolve this issue, set the open files limit to 4096 or higher by running the ulimit -n 4096 command.