Skip to content
Luke Gorrie edited this page May 6, 2015 · 18 revisions

Here are some tips for how to relate the software Architecture to nuts and bolts like files to install and init scripts to write.

Software installation

There is only one file to install: snabb. To create this file you can checkout the snabbswitch repository and run make at the top level to produce src/snabb. You can install this in a standard location such as /usr/bin/snabb.

snabb is a stand-alone executable that contains all of the required functions as sub-commands (in the style of busybox).

The relevant usages are:

The traffic command has no external software dependencies. Other commands expect certain programs to be in the PATH to call: snabb, git, mysqldump, diff, and standard shellutils.

What to run

These are the services that should run in addition to OpenStack itself:

  • neutron_sync_master: One instance per database server.
  • neutron_sync_agent: One instance per compute node.
  • traffic: One instance per 10G traffic port on each compute node.

The traffic process should run on a reserved CPU core (see Compute node requirements). The core can be assigned with taskset -c <core> snabb snabbnfv traffic .... Note: The command numactl seems to be incompatible with the Linux isolcpus feature and for this reason it is not recommended.

The command line arguments for each process can be determined by reading the usage and considering whether the default is suitable.

The processes print log messages to stdout and these should be redirected to a suitable location such as syslog.

Traffic process

Configuration

You do not have to write configuration files for the traffic application: these are generated automatically by neutron_sync_agent based on the Neutron database configuration.

Command line options

For the traffic process we recommend you consider using arguments similar to these:

  • --load-report-interval 5 for one-line summaries of traffic load every five seconds.
  • --link-report-interval 60 for a summary of internal traffic within the process every one minute.
  • --debug-report-interval 600 for a dump of the NIC hardware counter and status registers every 10 minutes.

traffic restarts

The traffic process is designed to be safe to restart. If it detects an error it will terminate with a message and expect to be restarted. Once restarted it will continue serving traffic for the virtual machines. (Stateful packet filtering connection tables are reset during restarts, however.)

We recommend you always restart the traffic process when it terminates, after some reasonable delay (e.g. between one and ten seconds).

Clone this wiki locally