NSO 6.7 is here!
Explore the highlights to discover what's new
Ctrlk
NSO DevCenterTry NSO

System Install

Install NSO for production use in a system-wide deployment.

Installation Steps

Complete the following activities in the given order to perform a System Install of NSO.

Prepare

1. Fulfill System Requirements 2. Download Installer/NEDs 3. Unpack the Installer

Install

4. Run the Installer

Finalize

5. Set up User Access 6. Set Environment Variables 7. Runtime Directory Creation 8. Generate License Token

Step 1 - Fulfill System Requirements

Start by setting up your system to install and run NSO.

To install NSO:

  1. Fulfill at least the primary requirements.

  2. If you intend to build and run NSO deployment examples, you also need to install additional applications listed under Additional Requirements.

Where requirements list a specific or higher version, there always exists a (small) possibility that a higher version introduces breaking changes. If in doubt whether the higher version is fully backwards compatible, always use the specific version.

Primary Requirements

Primary requirements to do a System Install include:

  • Linux for x86_64 architecture is required for production. Linux or macOS Darwin for x86_64 or macOS Darwin for arm64 is required for development purposes.

  • GNU libc 2.24 or higher.

  • Java JRE 17 or higher. Used by Cisco Smart Licensing.

  • Required and included with many Linux/macOS distributions:

    • tar command. Unpack the installer.

    • gzip command. Unpack the installer.

    • ssh-keygen command. Generate SSH host key.

    • openssl command. Generate self-signed certificates for HTTPS.

    • find command. Used to find out if all required libraries are available.

    • which command. Used by the NSO package manager.

    • libpam.so.0. Pluggable Authentication Module library.

    • libexpat.so.1. EXtensible Markup Language parsing library.

    • libz.so.1 version 1.2.7.1 or higher. Data compression library.

Additional Requirements

Additional requirements to, for example, build and run NSO production deployment examples, include:

  • Java JDK 17 or higher.

  • Ant 1.9.3 or higher.

  • Python 3.7 or higher.

  • Python Setuptools is required to build the Python API.

  • Often installed using the Python package installer pip:

    • Python Paramiko 2.2 or higher. To use netconf-console.

    • Python requests. Used by the RESTCONF demo scripts.

  • xsltproc command. Used by the support/ned-make-package-meta-data command to generate the package-meta-data.xml file.

  • One of the following web browsers is required for NSO GUI capabilities. The version must be supported by the vendor at the time of release.

    • Safari

    • Mozilla Firefox

    • Microsoft Edge

    • Google Chrome

  • OpenSSH client applications. For example ssh and scp commands.

  • cron. Run time-based tasks, such as logrotate.

  • logrotate. rotate, compress, and mail NSO and system logs.

  • rsyslog. pass NSO logs to a local syslog managed by rsyslogd and pass logs to a remote node.

  • systemd or init.d scripts to start and stop NSO.

Step 2 - Download the Installer and NEDs

To download the Cisco NSO installer and example NEDs:

  1. Go to the Cisco's official Software Download site.

  2. Search for the product "Network Services Orchestrator" and select the desired version.

  3. There are two versions of the NSO installer, i.e. for macOS and Linux systems. For System Install, choose the Linux OS version.

Identifying the Installer

You need to know your system specifications (Operating System and CPU architecture) to choose the appropriate NSO Installer.

NSO is delivered as an OS/CPU-specific signed self-extractable archive. The signed archive file has the pattern nso-VERSION.OS.ARCH.signed.bin that after signature verification extracts the nso-VERSION.OS.ARCH.installer.bin archive file, where:

  • VERSION is the NSO version to install.

  • OS is the Operating System (linux for all Linux distributions and darwin for macOS).

  • ARCH is the CPU architecture, for example, x86_64.

Step 3 - Unpack the Installer

If your downloaded file is a signed.bin file, it means that it has been digitally signed by Cisco, and upon execution, you will verify the signature and unpack the installer.bin.

If you only have installer.bin, skip to the next step.

To unpack the installer:

  1. In the terminal, list the binaries in the directory where you downloaded the installer, for example:

  2. Use the sh command to run the signed.bin to verify the certificate and extract the installer binary and other files. An example output is shown below.

  3. List the files to check if extraction was successful.

Description of Unpacked Files

The following contents are unpacked:

  • nso-VERSION.OS.ARCH.installer.bin: The NSO installer.

  • nso-VERSION.OS.ARCH.installer.bin.signature: Signature generated for the NSO image.

  • tailf.cer: An enclosed Cisco signed x.509 end-entity certificate containing the public key that is used to verify the signature.

  • README.signature: File with further details on the unpacked content and steps on how to run the signature verification program. To manually verify the signature, refer to the steps in this file.

  • cisco_x509_verify_release.py: Python program that can be used to verify the 3-tier x.509 certificate chain and signature.

NED Packages

The NED Packages that are available with the NSO Installation are netsim-based example NEDs. These NEDs are used for NSO examples only.

Fetch the latest production-grade NEDs from Cisco Software Download using the URLs provided on your NED license certificates.

Manual Pages

The installation program will unpack the NSO manual pages from the documentation archive in $NCS_DIR/man. ncsrc makes an addition to $MANPATH, allowing you to use the man command to view them. The manual pages are available in HTML and PDF formats and from the online documentation located on NCS man-pages, Volume 1 in Manual Pages.

Following is a list of a few of the installed manual pages:

  • ncs(1): Command to start and control the NSO daemon.

  • ncsc(1): NSO Yang compiler.

  • ncs_cli(1): Frontend to the NSO CLI engine.

  • ncs-netsim(1): Command to create and manipulate a simulated network.

  • ncs-setup(1): Command to create an initial NSO setup.

  • ncs.conf: NSO daemon configuration file format.

For example, to view the manual page describing the NSO configuration file you should type:

Apart from the manual pages, extensive information about command line options can be obtained by running ncs and ncsc with the --help (abbreviated -h) flag.

Installer Help

Run the sh nso-VERSION.linux.x86_64.installer.bin --help command to view additional help on running binaries. More details can be found in the ncs-installer(1) manual page included with NSO.

Notice the two options for --local-install or --system-install.

Step 4 - Run the Installer

To run the installer:

  1. Navigate to your Install Directory.

  2. Run the installer with the --system-install option to perform System Install. This option creates an Install of NSO that is suitable for production deployment.

    For example:

Default Directories

The System Install by default creates the following directories:

  • The Installation Directory is created in /opt/ncs, where the distribution is available.

  • The Configuration Directory is created in /etc/ncs, where the ncs.conf file, SSH keys, and WebUI certificates are created.

  • The Running Directory is created in /var/opt/ncs, where runtime state files, CDB database, and packages are created.

  • The Log Directory is created in /var/log/ncs, where the log files are populated. Also, init scripts are created in /etc/init.d/ncs and system-wide environment variables are created in /etc/profile.d/ncs.sh.

For the --system-install option, you can also choose a user-defined (non-default) Installation Directory, Configuration Directory, Running Directory, and Log Directory with --install-dir, --config-dir, --run-dir and --log-dir parameters, and specify that NSO should run as a different user than root with the --run-as-user parameter.

If you choose a non-default Installation Directory by using --install-dir, you need to specify --install-dir for subsequent installs and also for backup and restore.

For more information on ncs-installer, see the ncs-installer(1) man page.

For an extensive guide to NSO deployment, refer to Deployment Example.

Enable Strict Overcommit Accounting on the Host

By default, the Linux kernel allows overcommit of memory. However, memory overcommit produces an unexpected and unreliable environment for NSO because the Linux Out-Of-Memory (OOM) killer may terminate NSO without restarting it if the system is critically low on memory. Also, when the OOM killer terminates NSO, no system dump file will be produced, and the debug information will be lost. Thus, it is strongly recommended to enable strict overcommit accounting.

Heuristic Overcommit Mode as an Alternative to Strict Overcommit

The alternative—using heuristic overcommit mode (see below for best‑effort recommendations)—can be useful if the NSO host has severe memory limitations. For example, if RAM sizing for the NSO host did not take into account that the schema (from YANG models) is loaded into memory by NSO Python and Java packages affecting total committed memory (Committed_AS) and after considering the recommendations in CDB Stores the YANG Model Schema.

Recommended: Host Configured for Strict Overcommit

  • Set vm.overcommit_memory=2 to enable strict overcommit accounting.

  • Set vm.overcommit_ratio so the CommitLimit is approximately equal to physical RAM, with a 5% headroom for the kernel to reduce the risk of system-wide OOM conditions. E.g., 95% of RAM when no swap is present (recommended), or subtract 5 percentage points from the calculated ratio that neutralizes swap. Increase the headroom if the host runs additional services.

  • Alternatively, set vm.overcommit_kbytes which takes precedence; vm.overcommit_ratio is ignored while vm.overcommit_kbytes > 0.

    • When vm.overcommit_kbytes > 0, it sets a fixed CommitLimit in kB and ignores ratio and swap in the calculation. Note that HugeTLB is not subtracted when overcommit_kbytes is used (it’s a fixed value).

  • Strongly discourage swap use at runtime by setting vm.swappiness=1.

  • If swap must remain enabled system-wide, prevent NSO from using swap by configuring its cgroup with memory.swap.max=0 (cgroup v2).

  • If swap must be enabled for NSO use a fast disk, for example, an NVMe SSD.

Apply Immediately

When vm.overcommit_memory=2, the overcommit_ratio parameter defines the percentage of physical RAM that is available for commit.

The Linux kernel computes the CommitLimit:

CommitLimit = MemTotal × (overcommit_ratio / 100) + SwapTotal − total_huge_TLB

  • MemTotal is the total amount of RAM on the system.

  • overcommit_ratio is the value in /proc/sys/vm/overcommit_ratio .

  • SwapTotal is the amount of swap space. Can be 0.

  • total_huge_TLB is the amount of memory set aside for huge pages. Can be 0.

The default overcommit_ratio is 50%. On systems with more than 50% of RAM available, this default can underutilize physical memory.

Do not set vm.overcommit_ratio=100 as it includes all RAM plus all swap in the CommitLimit and leaves no headroom for the kernel. While swap increases the commit capacity, it is usually slow and should be avoided for NSO.

Compute overcommit_ratio to Neutralize Swap

To allocate physical RAM only in commit accounting and keep a 5-10% headroom for the kernel:

  • Compute the base ratio: base_ratio = 100 × (MemTotal − SwapTotal) / MemTotal.

  • Apply headroom: overcommit_ratio = floor(base_ratio) − 5.

Notes:

  • overcommit_ratio is an integer; round down for a bit of extra headroom.

  • Recompute the ratio if RAM or swap changes.

  • If SwapTotal ≥ MemTotal, swap cannot be neutralized via overcommit_ratio, use overcommit_kbytes; see Example 3.

  • If the computed value is very low, ensure it still fits your workload requirements.

Example 1: No Swap, 5% Headroom

Rationale: With no swap, set overcommit_ratio=95 to allow ~95% of RAM for user-space commit, leaving ~5% headroom for the kernel.

Example 2: MemTotal > SwapTotal, Neutralize Swap with 5% Headroom

Calculate the ratio:

  • base_ratio= 100 × ((8039352 − 1048572) / 8039352) ≈ 86.9%.

  • Apply 5% headroom: overcommit_ratio = floor(86.9) − 5 = 81.

This keeps the CommitLimit safely below physical RAM to provide kernel headroom and neutralizes swap’s contribution to CommitLimit and then applies 5% headroom toward the commit budget.

Example 3: SwapTotal ≥ MemTotal (Headroom via ratio not applicable, use overcommit_kbytes)

Compute:

  • CommitLimit_kB = floor(MemTotal × 0.95) = floor(16,000,000 × 0.95) = 15,200,000 kB.

Note that overcommit_kbytes sets a fixed CommitLimit that ignores swap; recompute if RAM changes. Also note the HugeTLB subtraction does not apply when using overcommit_kbytes (fixed commit budget).

Refer to the Linux proc_sys_vm(5) manual page for more details on the overcommit_memory, overcommit_ratio, and overcommit_kbytes parameters.

Persist Across Reboots

To ensure the overcommit remains disabled after reboot, add the three lines below to /etc/sysctl.conf (or a file under /etc/sysctl.d/).

See the Linux sysctl.conf(5) manual page for details.

NSO Crash Dumps

If NSO aborts due to failure to allocate memory, NSO will produce a system dump by default before aborting. When starting NSO from a non-root user, set the NCS_DUMP environment variable to point to a filename in a directory that the non-root user can access. The default setting is NCS_DUMP=ncs_crash.dump, where the file is written to the NSO run-time directory, typically NCS_RUN_DIR=/var/opt/ncs. If the user running NSO cannot write to the directory that the NCS_DUMP environment variable points to, generating the system dump file will fail, and the debug information will be lost.

Alternative: Heuristic Overcommit Mode

As an alternative to the recommended strict mode, vm.overcommit_memory=2, you can keep vm.overcommit_memory=0 on bare metal and use a best effort monitor to trigger NSO debug dumps before host-level memory pressure becomes critical. In heuristic overcommit mode, the most practical trigger for NSO debug dumps is a combination of low MemAvailable, sustained memory PSI pressure, and an elevated or clearly rising NSO oom_score.

  • This approach does not prevent NSO from getting killed; it attempts to capture diagnostic data before host memory pressure becomes critical and the Linux OOM-killer kills NSO.

  • Prefer low MemAvailable over MemFree, because MemAvailable is a better estimate of how much memory the host can still provide without heavy reclaim or swapping.

  • Use sustained PSI memory pressure from /proc/pressure/memory to confirm that the host is actively stalling on memory reclaim rather than reacting to a short-lived dip in available memory.

  • Use the NSO process oom_score from /proc/<pid>/oom_score to confirm that NSO is becoming a plausible OOM-killer victim. A high or clearly rising oom_score is more actionable than host memory metrics alone.

  • If swap is enabled, prefer vm.swappiness=1 and optionally include swap activity as additional context, but keep MemAvailable, memory PSI, and NSO oom_score as the primary debug dump trigger.

  • Tune the example thresholds for your workload. The values below are intentionally conservative starting points for bare-metal systems and should be validated under load.

  • Ensure the user running the monitor has permission to execute ncs --debug-dump and write to the chosen dump directory.

This recommendation also applies when NSO runs inside a Linux virtual machine. MemAvailable, memory PSI, and NSO oom_score remain the right guest-visible signals for predicting Linux OOM risk inside the VM. However, they reflect memory pressure seen by the guest OS and do not capture all hypervisor-level memory contention. If the virtualization platform reclaims memory from the guest or the host is overcommitted, guest performance can degrade due to host pressure even when the guest-side trigger has not yet fired. For production NSO virtual machines, avoid hypervisor-level memory overcommit for the VM when possible and monitor the virtualization platform's own memory reclamation signals in addition to the guest-side trigger.

Some older NSO releases expect the /etc/init.d/ folder to exist in the host operating system. If the folder does not exist, the installer may fail to successfully install NSO. A workaround that allows the installer to proceed is to create the folder manually, but the NSO process will not automatically start at boot.

Step 5 - Set Up User Access

The installation is configured for PAM authentication, with group assignment based on the OS group database (e.g. /etc/group file). Users that need access to NSO must belong to either the ncsadmin group (for unlimited access rights) or the ncsoper group (for minimal access rights).

To set up user access:

  1. To create the ncsadmin group, use the OS shell command:

  2. To create the ncsoper group, use the OS shell command:

  3. To add an existing user to one of these groups, use the OS shell command:

Step 6 - Set Environment Variables

To set environment variables:

  1. Change to Super User privileges.

  2. The installation program creates a shell script file in each NSO installation which sets the environment variables needed to run NSO. With the --system-install option, by default, these settings are set on the shell. To explicitly set the variables, source ncs.sh or ncs.csh depending on your shell type.

  3. Start NSO.

    Once you log on with the user that belongs to ncsadmin or ncsoper, you can directly access the CLI as shown below:

Step 7 - Runtime Directory Creation

As part of the System Install, the NSO daemon ncs is automatically started at boot time. You do not need to create a Runtime Directory for System Install.

Step 8 - Generate License Registration Token

To conclude the NSO installation, a license registration token must be created using a (CSSM) account. This is because NSO uses Cisco Smart Licensing to make it easy to deploy and manage NSO license entitlements. Login credentials to the Cisco Smart Software Manager (CSSM) account are provided by your Cisco contact and detailed instructions on how to create a registration token can be found in the Cisco Smart Licensing. General licensing information covering licensing models, how licensing works, usage compliance, etc., is covered in the Cisco Software Licensing Guide.

To generate a license registration token:

  1. When you have a token, start a Cisco CLI towards NSO and enter the token, for example:

    Upon successful registration, NSO automatically requests a license entitlement for its own instance and for the number of devices it orchestrates and their NED types. If development mode has been enabled, only development entitlement for the NSO instance itself is requested.

  2. Inspect the requested entitlements using the command show license all (or by inspecting the NSO daemon log). An example output is shown below.

Evaluation Period

If no registration token is provided, NSO enters a 90-day evaluation period and the remaining evaluation time is recorded hourly in the NSO daemon log:

Communication Send Error

During upgrades, If you experience a 'Communication Send Error' during license registration, restart the Smart Agent.

If You are Unable to Access Cisco Smart Software Manager

In a situation where the NSO instance has no direct access to the Cisco Smart Software Manager, one option is the Cisco Smart Software Manager Satellite which can be installed to manage software licenses on the premises. Install the satellite and use the command call-home destination address http <url:port> to point to the satellite.

Another option when direct access is not desired is to configure an HTTP or HTTPS proxy, e.g., smart-license smart-agent proxy url https://127.0.0.1:8080. If you plan to do this, take the note below regarding ignored CLI configurations into account:

If ncs.conf contains configuration for any of java-executable, java-options, override-url/url or proxy/url under the configure path /ncs-config/smart-license/smart-agent/, then any corresponding configuration done via the CLI is ignored.

License Registration in HA Mode

When configuring NSO in High Availability (HA) mode, the license registration token must be provided to the CLI running on the primary node. Read more about HA and node types in High Availability.

Licensing Log

Licensing activities are also logged in the NSO daemon log as described in Monitoring NSO. For example, a successful token registration results in the following log entry:

Check Registration Status

To check the registration status, use the command show license status.

System Install FAQs

Frequently Asked Questions (FAQs) about System Install.

Is there a dependency between the NSO Installation Directory and Runtime Directory?

No, there is no such dependency.

Do you need to source the ncsrc file before starting NSO?

No. By default, the environment variables are configured and set on the shell with System Install.

Can you start NSO from a directory, which is not a NSO runtime directory?

Yes.

Can you stop NSO from a directory, which is not a NSO runtime directory?

Yes.

For evaluation and development purposes, instead of a Local Install, you performed a System Install. Now you cannot build or run NSO examples as described in README files. How can you proceed further?

The easiest way is to Uninstall the System install using ncs-uninstall --all and do a Local Install from scratch.

Can we move NSO Installation from one folder to another ?

No.

PreviousLocal InstallNextPost-Install Actions

Last updated

Was this helpful?