A core port of the perfSONAR philosophy is test isolation: only 1 test should run on the host at a time. This
ensures that test results are not impacted by other tests. Otherwise it is much more difficult to interpret test results, which may vary due to host effects rather then network effects. For this reason, the heart of perfSONAR is the tool 'bwctl', which is designed to carefully schedule one test at a time.
This means you should not run perfSONAR tools on a host running a web server, a data server, or on a virtual machine, because in all of these cases it is not possible to guarantee test isolation.
As systems have gotten faster, and CPUs have more cores, we find that it is now safe to run a database, a.k.a. the perfSONAR Measurement Archive (MA), on the host. Inserts and queries to the database appear to have no impact on test results. Its also OK to run latency/loss tests (owamp) on one NIC, and throughput tests (iperf3/nuttcp) on a second NIC, without noticeable impact on the results. For more information on configuring a perfSONAR host for two NICs, see: http://docs.perfsonar.net/manage_dual_xface.html
Note: Running both owamp and iperf at the same time on the same NIC is a problem, and should be avoided. Sample results showing this are this page discussing perfSONAR Server Count.
For more information on hardware requirements, see Hardware Advice
perfSONAR on a Virtual Machine
There are many virtual machine technologies today, and while it is possible to deploy a VM with guaranteed CPU and NIC performance if you are using the right VM tools, it can be quite tricky to configure this correctly. Another item that is important to perfSONAR; accurate clock synchronization; is also harder to get right in a VM. For these reasons we recommend not using VMs for perfSONAR.
But if your only option is to run perfSONAR on a VM, you should do it, as there is still some value in the results. The most important thing to keep in mind is that your throughput results are likely to vary considerably, and will likely be about 50% lower than the physical host is capable of.
For more information on VM issues, see Virtual Machines
See Deployment Examples for configurations used by community members.
The perfSONAR Toolkit development team is soliciting configuration from the community at large. Examples of a configuration bundle would include:
- Version of software deployed
- Diagram of deployment locations
- Measurements configured
- Visualizations used (e.g. Maddash), including URLs if available
- Special configuration requirements (e.g. use of VLANs, modifications to testing software)
This information can be shared with the development team directly: email@example.com