Giving Measurement a Home
The perfSONAR node should be located based on simple criteria:
- Where it can be integrated into the facility software/hardware management systems?
- Where it can do the most good for the network operators or users?
- Where it can do the most good for the community?
Always locate the perfSONAR node in a location that has proper HVAC and power considerations. Most times this means a network or computing rack.
In some circumstances a perfSONAR node may not be server-class hardware, and could be placed in other locations, e.g. an office, near a wireless controller, or other temporary location as problems are debugged.
General Considerations for Node Placement:
Operating a perfSONAR node should be done in a consistent manner, based on how other servers within the domain are administered. This implies several considerations:
- The server must become a citizen of the network. This means it has to be treated the same as other servers. If your site uses configuration management software, install the same software on the perfSONAR node (see this page for more information). If the other servers use two factor authentication for login, or a jump host that restricts SSH sources, integrate these features. The perfSONAR node cannot become an orphan, otherwise the risk of compromise increases.
- If the perfSONAR node is treated as disposable, e.g. it is a test beacon only that does not contain any PII (Personally Identifiable Information) and is not integrated into the campus, take precautions to protect the host as an untrusted member of the community. For instance, do not recycle common passwords or store keys on the machine. Lock down access to key services (e.g. SSH) so they can only occur from designated locations, and use the perfSONAR features that make regular data backups.
- In either deployment case, consider the following suggestions for administration:
- Restrict SSH logins to within the domain
- Maintain IPTables rules to restrict ports
- Enable automatic updating of yum
- Configure rsyslog to send logs to an external host or analyzer
- Install additional IDS software, beyond the defaults for the perfSONAR Toolkit
- If applicable, use a management network on a spare networking interface to further restrict the access profile for users and system data.
- Give the node a descriptive name. E.g. If its a border tester, or located in a Science DMZ, consider placing that element into the host name. IF the host is 10G, or 40G, consider using this as well. An example: perfsonar-dmz-10g.school.edu.
perfSONAR nodes are only as secure and useful as the local site can make them. Practicing accepted system administrator policies is recommended to ensure machines are not compromised or suffer from bit-rot.
Local Placement Considerations
Local placement, either nodes that are configured to use an RFC1918 address or those that are public, should abide by the following considerations:
- Connect the node to the local networking infrastructure in a consistent manner with other devices. For example, if the perfSONAR node is located in the same facility as data movement machines, utilize the same switch/routing infrastructure. This will give the perfSONAR node the same view of the outbound network path. Note that if the access/top-of-rack switch is not capable of handling the load of the data movement machines (e.g. lack of buffering, congestion because of high fan-in), the perfSONAR nodes will react and produce data consistent with a problem.
- Place nodes where they will be most useful. If the facility is distributed, a node at the demarcation of each location (near a building router/switch) can help detect problems with the MAN infrastructure. A perfSONAR node located adjacent to functions (processing, storage, or scientific instruments) can similarly find problems with the portions of the network path that are close to these devices.
If a facility is interested in collecting internal data, be aware of several properties of network measurements over short paths:
- Throughput tests are unlikely to be useful, as short RTT tests will frequently exhibit high values, even if packet loss or other factors are impacting TCP.
- Throughput tests will consume network resources quickly, and due to the short RTT may not back off; this could hurt background traffic that is trying to use network resources and traveling over a longer RTT.
- Latency testing from OWAMP requires stable, accurate, and precise measurements from NTP. In many cases NTP can synchronize a clock to the microsecond level, in others there may be a few milliseconds of difference. For short RTTs, the clock skew will become very noticeable, and in some cases may produce negative latency values due to the error in NTP correlation.
Packet loss measurements (also delivered via OWAMP) are the most valuable local measurement to consider.
For more information about local placement see also Local Deployments
Global Placement Considerations
If the perfSONAR node is advertised to the public (e.g. it has a public facing IPv4 or IPv6 address), external parties may use it for testing. There are several considerations to abide by when exposing a node for public use:
- Machines located on the border of a site, in front of any firewalls, will give external parties a good indication of the performance to the demarcation of your network.
- Machines located inside of a site border, with firewalls in the path, will give external parties a good indication of performance within your network. Note that this is vastly different than the previous case, and could give different results.
If your facility has centralized computing, storage, research activity, or a Science DMZ, it is recommended that a perfSONAR node be placed here as a service to external users that may use these facilities. This location would be paramount for supporting these users, and isolating end to end problems. Locating a machine at the border is more valuable from a networking perspective, to test the SLA agreements that are in place with WAN providers.