Barefoot Networks Launches Deep Insight
SANTA CLARA, Calif. -- Barefoot Networks today announced Barefoot Deep Insight, the world’s first network monitoring system to provide full visibility into every packet in a network. Barefoot Deep Insight software, running on commodity servers, interprets, analyzes and pinpoints a myriad of conditions that can impede packet flow, and does so in real time and at line-rate. An intelligent and programmable triggering mechanism allows only the interesting network events to be detected and reported in real time. Irrelevant data is automatically filtered out. Machine learning can be employed to achieve stateful baselining of the network performance and automatic detection of anomalies at any time scale and with nanosecond resolution.
In networks, today, even the very simplest questions regarding visibility are impossible to answer. For example, if we examine a data packet in the network and ask, “How did this packet get here?” and “How long was it delayed at each switch, router and firewall along the way?”, there is no way to tell. This is because most networks still use the same tools as 20 years ago: traceroute, ping, SNMP and NetFlow. At best, we can find statistics based on samples. This isn’t enough to capture a transient event, such as a microburst, and track down its root cause.
A breakthrough in network monitoring, Barefoot Deep Insight is first to provide the four ground truths for every packet in the network:
(1) How did it get here? The sequence of switches the packet visited along its path.
(2) Why is it here? The set of rules it matched upon at every switch along the way.
(3) How long was it delayed? The time it spent buffered in every switch, to the nanosecond.
(4) Why was it delayed? The packets, flows and applications that the packet shared each queue with.
Armed with this information, Deep Insight detects almost any anomaly in the network. For example, it can detect microbursts, however short, and answer all the above questions about every packet in the microburst. It can identify which application is causing congestion, whether a packet is in a part of the network it doesn’t belong, and if an ECMP is not load-balancing equally. It can even present the operator with individual dropped packets, and say why they were dropped. All of this at line-rate and multi Tb/s throughput without extra hardware, or extra packets. Anomalies are typically detected within 100 microseconds, yet can be captured and post-processed for up to a month. Deep Insight has been successfully evaluated and is production qualified at large customers including ones in cloud and enterprise markets.