Highly accurate latency measurements for data center networks

Data center operators serving customers that provide services for low-latency-critical applications such as electronic trading, multi-player gaming, Voice over Internet Protocol or video conferencing are under pressure to measure latency and loss in their networks and existing tools to make such measurements do not provide adequate granularity, according to a group of computer scientists from Purdue University and University of California, San Diego.

In a recently published paper, four scientists from the two universities proposed a new solution that they say will provide latency and loss measurements at extremely small scales – down to tens of microseconds. The mechanism, called Lossy Difference Aggregator, can be installed either within a router or across multiple routers to assist with fault localization.

Neither of the mechanisms current routers support – SNMP and NetFlow – “are up to the task,” the paper’s authors wrote. SNMP’s cumulative counters can provide loss estimation but cannot estimate latency and the procedure required for NetFlow to calculate latency demands “fundamentally high space complexity.” Running NetFlow at high-enough sampling rates usually requires network throughput that cannot be accommodated due a system’s operational demands.

Because of SNMP’s and NetFlow’s shortcomings, operators of latency-critical networks use external monitoring solutions that often lead to “grossly inaccurate” latency estimates.

The proposed LDA solution “accurately measures loss and delay over short time scales, while providing strong bounds on its estimates, enabling operators to detect short-term deviations from long-term means within arbitrary confidence levels.” The mechanism requires very low bandwidth overhead, is customizable and can help with fault localization.

Read the paper, written by Ramana Kompella, Kirill Levchenko, Alex Snoeren and George Varghese, here.

Related news: New software aims to turn data centers into single plug-and-play networks
Related feature: Copper and fiber decisions for 10 Gb/s data center
Related video: Web 2.0 and the impacts/demands of network hardware in your datacenter