Lawrence Joint Techs (JT12) Summary

By Ben Chinowsky

The 12th Joint Techs Workshop (JT12), cosponsored by ESCC and Internet2, took place August 3-7 at The University of Kansas. Some major themes were:

• Optical networking. There have been major developments since this was featured at JT11. National LambdaRail (NLR) is launching an experimental optical infrastructure (as vs. the production optical infrastructure of Abilene). The first NLR links are expected to be up this fall. FiberCo is a spinoff from NLR for bulk purchase and allocation of fiber in support of regional initiatives; FiberCo Technical Director Paul Love's presentation has a list of these initiatives. A Practical Optical Networking session offered insights on lots of new physical-layer issues to contend with. Fiber is finicky ("always clean fiber three times before plugging it back in" was one lesson from the CENIC fiber build) and can be dangerous, and the huts where gear is stored are often hard to get to, both geographically and administratively. Jerry Sobieski, Director of Network Engineering for the MAX gigaPoP, strongly recommends Jeff Hecht's Understanding Fiber Optics for background.
   
• Continuing the network buildout. Research and education networks continue to grow apace; for example, Jim Leighton noted that ESnet has sustained 100% growth per year since 1990. Rick Summerhill reported that the Abilene upgrade has gone "as smoothly as one could ever hope for," and is almost done. Summerhill expects that 1 Gbps will become the normal expectation of researchers this year. On the campuses, continued rapid growth is creating more and more issues at the sub-physical layer (what the leading champion of proper attention to these issues, Minnesota's Dave Farmer, calls "Layer 0"), which involves things like digging trenches and finding closet space. Expect more on Layer 0 issues at JT13.
   
• Security. Kent Pope offered a compelling overview of just how bad things are getting, and a vision of active net defense. Bill Manning gave a comprehensive introduction to DNSSEC, taken from his still more comprehensive introduction to DNS in general. Denial of Service (DoS) and — above all — Distributed Denial of Service (DDoS) attacks are emerging as especially difficult threats to cope with, and were a major focus of security-related sessions throughout JT12. A special double-length in-depth session was devoted to DDoS. The human-factors and political issues involved in troubleshooting DDoS attacks are huge: attacks may come from or through many other networks, including those of competitors, and thwarting these attacks can involve a tremendous amount of work for network engineers and inconvenience for users. Automatic mitigation is the Holy Grail of response to DDoS, but this an awful lot of responsibility to ask a network engineer to turn over to a computer. Presentations from the DDoS workshop are at http://mypage.iu.edu/~daripley/lawrence/.
   
• Network transports. The Internet2 End-to-End Performance Initiative (E2Epi) has made a priority of resolving the problem of packet loss due to causes other than congestion. This in turn has many causes, but three stand out: physical-layer problems (e.g., dirty fiber, crummy cabling), duplex mismatch, and inadequate network transport. Of these, the first two can mostly be solved by applying what we already know. Transport is much harder — how to tune, modify, or replace protocols designed for kilobits, to cope with gigabits? As a step toward an answer to this question, presenters shared their experiences with four advanced transport protocols: FAST, SABUL, Tsunami, and XCP. FAST developer Steven Low described advanced transports and large MTUs (which were the subject of much discussion at JT11) as complementary approaches to end-to-end performance. Detailed summaries of several E2Epi sessions are available on the E2Epi web site.

  Looking further down the road, Juniper's Dennis Ferguson argued that ultimately a better solution would be to replace TCP with end-to-end optical circuit switching, or reserved bandwidth more broadly, via "a more generic host bandwidth reservation protocol, which didn’t depend on the existence of end-to-end optical switching (but could use it when available)". On the other hand, John Golub, telepresent from Jerusalem, gave a sobering account of developments in optical switching over the last several years; with billions of dollars spent, there is "essentially no product presence in the market today".

• Network measurement. This was another aspect of end-to-end performance improvement that received a lot of attention at JT12. As discussed in depth at a measurement workshop following JT11, if there is to be optimal end-to-end performance there must first be end-to-end measurement. This in turn necessitates the construction of Network Measurement Infrastructures. Warren Matthews described the use of PingER to measure the global digital divide; surprisingly, this work revealed only weak correlations between Human Development Index, GDP, and education on the one hand, and network performance on the other.

As usual at JT, there were also many other developments discussed that didn't fit into any of the above categories. Internet2 Chief Engineer Guy Almes noted that there is keen interest in cyberinfrastructure from a wide range of disciplines; the NSF cyberinfrastructure report is available online. A report from the CMU Bandwidth Management project showed the amazing results that can be achieved by talking to the users. ARIN President and CEO Ray Plzak solicited the R&E networking community's input on address-allocations issues related to IPv6 multihoming; see http://www.arin.net/policy/discussion.html.

Presentations from JT12 are available at http://www.ncne.nlanr.net/training/techs/2003/0803/agenda.html.