« Metrics, the network operator and the network user | Main | Vacation Area Network »
March 22, 2003
Open-source network measurement
My recent focus/interest has been network measurement. This is partly motivated by a requirement for an IP QoS solution to allow us to converge our video, voice and data networks at UEN. But it started shortly after I joined UEN, when Jim and I started talking about what kinds of metrics a network operator can be and should be measured by, and that would drive the right kind of technical, operational and organizational changes to make the network better.
At NGN last fall, Jim and I attended a great tutorial on application performance measurement. This started a focus to find ways to better measure users' experience on our network. We've found some metrics that help identify obvious problems, such as saturated circuits and overloaded routers, but there are so many performance issues that don't show up in the metrics we can currently measure.
Vendors like Brix, NetScout, NetQoS have systems to passively or actively measure performance of certain applications across the network. These look like great solutions and are successfully used by several (maybe many) networks. The latest BCR magazine has reviews of these and other similar (commercial) tools. They are all quite expensive, and we would only be able to deploy them in a limited fashion on our network. So I've spent the last few weeks looking for other options, especially open-source ones. I'm impressed.
iperf is a measurement tool developed by NLANR. It's been used very extensively in the research community for projects like Web100 and the Abilene network. iperf can measure node-to-node throughput, packet loss, jitter and latency. It has a wide variety of options to measure all sorts of network characteristics. It runs on Linux, Solaris, Windows, MacOSX and any other platform that can compile the source-code. iperf (coupled with MRTG or Cricket for graphing) looks like a great possibility to actively measure the performance of different kinds of traffic across the network, and because it is portable and inexpensive (free), it could easily be deployed pervasively throughout the network. At least functionally, this would do the same thing as the commercial tools offer, but not as beautifully.
PastTmon is an Application Response Time monitor. It passively watches TCP traffic flows and records the performance of the application using PostgreSQL. It has a nice interface using PHP and the R graphing tool. Though PasTmon is early in development, it already looks very useful, and be able to provide the information we need to monitor TCP application performance. As an early user of the product, we may be able to make it more successful through our feedback and direct contributions to the project.
I think it would be interesting to try a Web Services approach to integrating these tools with the rest of our network management tools. These tools could also be extended to support more direct monitoring of applications, where a Web Services approach would also be very powerful.
These options probably pale in comparison to the commercial solutions we've investigated. But I have been impressed with the capabilities of an free/open-source solution, and I think it can meet the majority of our needs for measuring user-relevant metrics. Given the costs of the commercial tools, the capabilities of the free/open-source tools, and the budget constraints most organizations are dealing with, the free/open-source tools present a pretty compelling option financially and functionally.
Posted by pete at March 22, 2003 01:24 PM
Trackback Pings
TrackBack URL for this entry: