Passive Monitor: equinix-sanjose

The equinix-sanjose Internet data collection monitor is located at an Equinix datacenter in San Jose, CA, and is connected to an OC192 backbone link of a Tier1 ISP between San Jose, CA and Los Angeles, CA.

Hardware

The infrastructure consists of 2 physical machines, numbered 1 and 2. Both machines have a single Endace 6.2 DAG network monitoring card. A single DAG card is connected to a single direction of the bi-directional backbone link. The directions have been labeled A and B. Both machines have 2 Intel Dual-Core Xeon 3.00GHz CPUs, with 8 GB of memory and 1.3 TB of RAID5 data disk, running Linux 2.6.15 and DAG software version dag-2.5.7.1. In a test environment both machines dropped less then 1% of packets with snaplen 48 at 100% OC192 line utilization, using a Spirent X/4000 packet generator sending packets with a quadmodal distribution, with peaks at 40, 576, 1500 and 4283 bytes.

Realtime Traffic Report Generator

Realtime traffic reporting for this monitor is available at

http://www.caida.org/data/realtime/passive/?monitor=equinix-sanjose-dirA (direction A)
http://www.caida.org/data/realtime/passive/?monitor=equinix-sanjose-dirB (direction B)

Since at OC192 line rates this hardware can't keep up with generating flow files that are needed for the realtime traffic reports, the reports are generated using flow estimation described in the paper Building a Better Netflow. The flow estimation tools that are used are part of the CoralReef suite of tools. Specifically, crl_anf is used with the following command line parameters:

crl_anf -b -A 500000 -f 500000 -Calignint -O %flow_file% -Cm=phy=pos %measurement_card%

The -b option outputs a binary format, for increased efficiency in writing and storage. The -A and -f options set the maximum number of entries in the tables for counting using the ANF and FCE algorithms, respectively. (Note: as of April 20th, 2008, this functionality is not yet part of a CoralReef release, as it has not been fully tested and documented.) -Calignint forces the (5-minute) intervals to align along 5-minute clock boundaries, so as to make it easier to compare intervals from different monitors, and -O specifies the output filename. The rest is for configuring the DAG card.

Traffic traces

The first passive traffic traces on this monitor were taken July 2008 and are part of the CAIDA Anonymized 2008 Internet Traces Dataset.

Time Synchronization

Both physical machines are configured to synchronize their hardware clock via NTP. The DAG measurement cards have their own internal high-precision clock, that allows it to timestamp packets with 15 nanosecond precision. Every time a traffic trace is taken, The DAG internal clock gets synchronized to the host hardware clocks right before measurement starts. NTP accuracy is typically in the millisecond range, so at initialization the clocks on the individual DAG measurement cards can be off by a couple of milliseconds relative to each other. The precision (ie. timing within the packet trace) within a single direction of trace data is 15 nanosecond for the DAG files, and 1 microsecond for PCAP files. Due to human error the DAG internal clock was not synchronized with the system hardware clock for the first passive traffic trace (July 2008).

Starting in August 2008 the clocks on the DAG measurement cards are synchronized between the two measurement machines. At the beginning of each measurement experiment, the DAG clocks are synchronized to the individual machine clocks once, but one DAG card also feeds a 1 pulse per second signal to the other card, which allows them to be synchronized up with approximately 2 clock ticks precision (30 nanoseconds).