Router tool tracks network delays down to microsecs

Researchers at Purdue University and the University of California have proposed a software enhancement for IP routers which, they say, will enable them to measure microsecond-level latency fluctuations that can beset high-performance applications such as automated trading, videoconferencing, and online gaming.

Although there can be multiple contributory causes for latency loss, errors recurring within routers due to bugs or misconfigurations have been costly to find and fix.

The Lossy Difference Aggregator (LDA) is a measurement data structure that measures packet delay and standard delay deviation, and can help with fault localisation. It complements conventional probes used to monitor-end-to-end applications performance, its designers say. Compared to alternative solutions, LDA is inexpensive to implement on existing devices, the researchers say, has a low processing and bandwidth-usage overhead, and takes advantage of unallocated transistors on next-generation 65-nm router chips.

Operators of latency-critical networks are forced to use external monitoring mechanisms to collect sufficient number of samples to compute accurate estimates, the LDA researchers claim. Capturing these effects in live networks requires injecting a prohibitively high rate of probe packets; so operators employ external passive hardware monitors at key points in the network, which can be an expensive exercise.

The LDA aims at tracking packets that go awry within a router – between ingress and egress – rather than during transmission between routing devices. In general it works by grouping packets into ‘measurement segments’ as they pass through a router and ‘decomposing’ their incoming and outgoing sum totals to check for fluctuations. This way specific routers that might be causing or contributing to latency delays, can be highlighted and debugged.

The LDA research was authored by Ramana Rao Kompella, Kirill Levchenko, Alex C. Snoeren, and George Varghese; it was supported by the National Science Foundation, and Cisco Systems.

Further information:
http://cseweb.ucsd.edu/~snoeren/papers/lda-sigcomm09.pdf

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close