This is the first OC48c Packet-over-SONET data set published by the NLANR MNA team. It consists of a pair of two hour contiguous bidirectional packet header traces collected at the Indianapolis router node (IPLS), which is (at the time of collection) a CISCO GSR 12015 with four OC48c uplinks, four OC12c links and one OC3c link (a mixture of ATM and Packet-over-SONET, see OC48c backbone map below). The links instrumented are the ones eastbound and westbound, towards Cleveland (CLEV) and Kansas City (KSCY).

The initial plans were drawn for a complete router instrumentation (all of the links) in March 2001. Funding was available in May 2001, but then the project was stalling due to the unavailability of OC48c monitoring gear, which eventually became available in May 2002 in the form of the Dag4.23 cards from Endace Measurement Systems.

The installation itself proved very challenging and eventually the decision was made to focus on just a pair of OC48c links. The original goal to provide packet delay and loss figures across a busy backbone router is still met to a certain degree.

The data on IPLS-CLEV is time-correlated to the IPLS-KSCY data via the DUCK timestamping system, the precision of which has been recorded (see below). Both monitors use the feed from a Praecis Ct CDMA time receiver, the 1PPS signal of which is distributed via an Endace TDS-24 system to all of the cards in the measurement configuration. The same anonymization has been applied to both sets of trace files and packets can be matched based on the information visible in the trace files (the first 44 bytes of the IP header, anonymized).

We are currently preparing a set of delay graphs based on the WAND dagtools.

There is no question that a project like this one depended on the contribution and support of a long list of a great many individuals.

To begin with, strong support was provided by the National Science Foundation, our funding agency, in particular by Chip Cox, Tom Greene and Greg Monaco.

Matt Zekauskas, Advanced Networks and Services and Internet2, has been a great supporter troughout the project. With Matt, we inspected the NYCM and IPLS router nodes on different occasions, and Matt helped steering through the difficult times of getting the installation organized and the MOU signed. We also jointly tested CDMA time receivers and found them to be working reliably in colocation areas, which is a breakthrough for the network research community that relies on accurate timing. Thank you, Matt!

This project would have been doomed without the support of Steve Corbató, Director Backbone Infrastructure at Internet2. We are grateful to Steve for permission to tap into the core infrastructure, to appreciate the importance of this project for the research community and to back the sophisticated installation work that we carried out at the Qwest POP. Steve, we are looking forward to working with you on the next project!

At critical moments during the preparation of this installation, we could rely on Guy T Almes, Chief Engineer at Internet2 and Advanced Networks and Services. We can do this together, indeed!

In many, many situations John Hicks, TransPAC/Indiana University, was the first one to be there with support at all fronts. John organized the first visit to the POP in 2001, he was there to receive the equipment as it got shipped, he hauled the gear trough the campus, across the floors, up and down the staircases, he was available all the three days (and nights!) we went to actually install the equipment at the POP. John's wife Julie even took care of Phillipp (11), my son, during our stay at Indianapolis. If you ever needed a partner-collaborator on a research project, John, without a doubt, would be the ideal. Thank you, John!

The Abilene Team at Indiana has simply been marvellous. It has been a great pleasure to work with Chris Robb, Grover Browning and Caroline Carver. We appreciate the late night hours you spent with us in the POP to make this project happen. Thanks to all of you.

The OC48MON was originally a research project by CAIDA and the development of the Dag4 series of passive measurement cards was outsourced to a team led by Professor Ian Graham as part of the WAND research group at the University of Waikato in New Zealand. Three years in the trenches of fighting with high-speed hardware have finally payed off, and we are relieved and proud of the result.

Naturally, everyone within the NLANR MNA team has suffered through the 18 month of making this project happen and deserves kudos in many ways. However, none of those traces would be available here today for download, if it was not for the magic of Hans-Werner Braun. He can resolve problems that appear hopeless. Trust me.

The traces published are in Dag PoS format, which is a 64 byte record per packet, starting with a 64 bit timestamp, some overhead bytes and a total of 44 bytes from the beginning of the IP header. Documentation about this format can be found here. We recommend the WAND dagtools for parsing the data. The dagdump utility with option -p will deliver an ASCII printout, dagbpf -p converts the file into libpcap/tcpdump. Notice the issues regarding the wlen field below.

Traces are available both via FTP and HTTP:

• http://pma.nlanr.net/Traces/Traces/long/ipls/1/
• ftp://pma.nlanr.net/traces/long/ipls/1/

MD5 checksums for all trace files are here.

The timestamping precision has been recorded: IPLS-CLEV and IPLS-KSCY.

The -0 direction is westbound, -1 is eastbound, for both sets of data.

Late breaking news. After recording the trace files we figured there are problems with the wlen field, which are related to configurations of the Vitesse 9112 OC48c physical layer interface chip. We are preparing a report on that went wrong, and how to correct for those errors. In the mean time, for all of your analysis works, please rely on the IP total length field and stay away from analysis using the wlen field. There are off-by-one, off-by-two as well as off-by-four errors.