Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Configuring Enhanced Orchestration and Hugepages

Enhanced Orchestration

Enhanced orchestration mode enables you to easily manage VNFs and service chains without requiring the VNF XML descriptor files. By default, this mode is ON and this is the recommended mode.

To enable enhanced orchestration:

Note:

Ensure that you reboot the system after enabling the enhanced orchestration mode.

Hugepages

Hugepages in NFX250 devices are contiguous memory blocks of 2 MB and 1 GB sizes, which are used for virtual memory management. Due to memory fragmentation, the system might not have sufficient memory to assign the required amount of hugepages when a new VNF is launched. This might result in the VNF either failing to launch or switching to the shutdown state during configuration. Hence, to launch VNFs on NFX250 devices, you must pre-allocate memory using hugepages before configuring the VNFs.

Note:

Hugepages must be enabled for all VNFs that use OVS for service chaining.

When pre-allocating hugepages, ensure that there is sufficient memory for system use as insufficient memory might cause the system to become unresponsive. The system, which consists of the JCP, JDM, and Hypervisor, requires around 6 to 7 GB of memory. Only the remaining memory can be used by the VNFs.

To view the current state of memory and hugepages available, issue the following command at the JDM CLI prompt:

The default configuration comes with one hugepage of 1 GB size that is used by the OVS, and hugepages of 2 MB size for system use.

Table 1 lists the maximum hugepage memory that can be reserved for the various NFX250 models.

Table 1: Recommended Hugepage Memory for the NFX250 Devices

Model

Memory

Maximum Hugepage Memory (GB)

Maximum Hugepage Memory (GB) for CSO-SDWAN

NFX250-S1

16 GB

8

-

NFX250-S2, NFX250-S1E

32 GB

24

13

NFX250-LS1

16 GB

8

-

Pre-allocating Hugepages

To pre-allocate hugepages, issue the following command at the JDM CLI prompt:

where:

  • page-size can be one of the following values:

    • 1024 for 1 GB hugepage

    • 2 for 2 MB hugepage

  • page-count is the number of hugepages

The page-size and page-count values depend on the size and total number of hugepages required by all the VNFs that will be launched in the system. The values are also limited by the available memory on the device.

We recommend that you reboot the system after configuring hugepages to pre-allocate hugepages during bootup.

Note:

By default, OVS uses 1 GB hugepage from the allocated set of hugepages.

The following sample output shows a configuration with both 1 GB and 2 MB hugepages configured. A total of 8 hugepages of 1 GB size are configured, in which one hugepage is used by the OVS and the remaining seven hugepages can be used for the VNFs. There are 376 hugepages of 2 MB size that can be used for VNFs. The remaining hugepages (500-376=124) are reserved for system use.

Note that the value of Configured 1 GB Huge Pages includes the hugepage that the system allocates for the OVS.

Note:

On NFX250 NG devices, the value of the Total 1GiB Huge Pages will be one more than the total hugepages reserved through the CLI (Configured 1GiB Huge Pages).

Allocating Hugepages for a VNF

To allocate hugepages for a VNF, issue the following command:

Troubleshooting Hugepages

If the number of hugepages configured results in insufficient memory for system use, the following error message appears when you access the JDM CLI after rebooting the device:

fatal error - could not reserve address space in "getmem.c"

To reconfigure the hugepages:

  1. Establish an SSH connection to the hypervisor:

    user@jdm# ssh hypervisor

  2. Determine the number of hugepages configured:

    cat /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages

  3. Reset the number of hugepages to 0:

    echo 0 > /sys/kernel/mm/hugepages/hugepages-1048576kB/nr_hugepages

  4. Reboot the device.

  5. Access the JDM CLI and reconfigure the number of hugepages.