Surveyor Site Contacts Meeting

October 19, 1998

Introduction

There were about 25 attendees. The technical expertise appeared very high, about 6 came from Advanced.

Current deployment/infrastructure (Matt Zerkauskas)

First sites May 1997 (Advanced & UChicago)

Can see asymmetries with one-way. They have long term collection with trends & post-mortems, they have started correlating (since end of last year) with traceroute so they have a path history.

Potential users: network operators, network engineers, end-users, application developers, and network researchers. Advanced want to solve the limited problem of academic network performance.

Now have surveyors in Canada, Alaska, Hawaii (DREN), New Zealand, working with France Telecom (Renater), CERN (has received equipment), possibly DANTE, UUNET (in Norway). The current deployment is 33 sites, 893 paths, CSG 15, National Labs 8, I2 GigaPoPs, tele-immersion 5. They are being very aggressive in deployment. They are looking for how things can be improved.

Interesting network observations - Sunil Kalidindi

Basic surveyor is a 400MHz Dell Pentium. They have about 200GB fiber atached RAID, about half full. 40 byte pings 12 bytes of data. They have a GPS on each surveyor and will be putting the lat/long into the DNS database.

Basic reporting tools are the daily summary reports of loss & delay. They collect data every 2 minutes. They report 3 statistics: minimum, 50% and 90%. Everything is UTC. There are 1440 X values per day (i.e. 1/minute) from measurements made 2/second. There are often small gaps in the measurement due to lack of a GPS clock at either site. they have thought about trying to extrapolate the GPS clock when it gets lost. Most GPS clock uptimes are over 90%.

Often see asymmetries even on symmetric routes. For example Univ Hawaii & SLAC, Univ. Hawaii to Univ. Michigan has a symmetric route and minimum delays similar, yet the delays in the 2 directions are very different.

NASA-Ames to Univ Aukland is terrible with big packet delays, due to chronic lack of capacity. It is much better in the reverse direction. As a result Email & Web work OK, interactive is not possible, and they are experts in Web caching.

The scatter in the delay changed dramatically for U Virginia to U Washington on 24th September 1998, even though the minimum delay slightly increased (vBNS is a fairly sparse geographic backbone so the path-length may be longer than the commercial Internet).

They showed an example of FNAl to U Hawaii there was a dramatic 57ms to 140ms delay in the minimum delay time on Oct 9/10, 1998. Yet the routes were identical according to traceroute.

They are beginning to need alerts to be sent to campuses. Looking through 900 paths is not practical. They are interested in sending email to site contacts when they detect abnormalities.

Miscellaneous

Kelly Deyoe (kdeyoe@cmu.edu) from CMU showed a Lucent developed 802.11 standard compatible 2Mbps PCMCIA wireless LAN card. It is not available yet, though is appearing in advertizing literature. It is expected to cost about $300 and works with a wireless Access Point that he estimated will cost on the order of $1K. The Access Point has a 10Mbps uplink. It works within buildings, so it would work well for conference rooms. It has 10 channels to one can have more than one in the same building (on different chanels so they do not interfere). If you are roaming it will hand-off as one moves from one Access Point to another. Typically they expect a couple Access Points per floor.

The Internet access was by way of a Bay Networks Netgear 56kbps Remote Access Modem Router. It has a built in 56kbps modem for the uplink, and 4 10Mbps Ethernet ports. It costs about $279 and seemed to work very well for connecting people in the meeting.

DataBase

Used to be Oracle, but observation was that it was resource intensive (money, disk, cpu & people skills) and did not provide much added value since they weren't using the relational features. They now use a flat file and store by path and day. They save IP address only (not name - this may be a problem later), they have 8 binary bytes/delay measurement (the time stamp is the number of usec since the beginning of the day) with a fixed layout to simplify searching. It is compressible by only 25% so it is already pretty efficient.

User Interface

They provide interfaces to view by date or by path (the path provides a map to help in the selection). Then they allow drilling down to a particular a particular path for a particular metric for a particular day. They also have an animation feature which is entertaining and may be useful for looking at trends or similarities (e.g. each day there are longer delays at 2am when the network is busy doing backup). They do not seem to have long term summary plots yet.

To-do Wishlist from a year ago

NTP support from Surveyor Measurement machines. They undersatnd what needs to be done, but do not have the resources to do this work. Is anyone interested? It is not a critical part of the Internet measurement process, but would be a nice added value.

Fiber antennas. Worried about lightening, fiber antennas exist but are expensive ($10K).

Cable & connectors. They now provide connectors and cabling with each new deployment.

Installation FAQs. No new status, they have found the exisiting documents sufficient.

Descriptions of various DMZs. They are still asking new sites for this information, but it has not been organized or made available outside Advanced. There are concerns about security issues of making this openly available. This might provide all the offsite connections, whether they are using ATM, what is in between the Surveyor and the outside world (e.g. switches, routers, speeds etc.)

ATM. This awaits primarily upon BSDI ATM drivers. The latest release of BSDI UNIX (version 4.0) does not include ATM drivers.

Multiple connections. No new status.

UDP request to perform traceroute. They are conducting traceroutes for each path according to a Poisson distribution which does not exceed 10 minute intervals.

Adaptive traceroute. No new status.

TTL measurement/record. No new status.

Time zone adjustment for graphs. No new status