| ||||||||
Status Report - June 2003
Activities extending the Network Analysis Infrastructure (NAI) in support of new and developing HPC needs:A series of graphs that plot delay statistics for the previous 8 months of both IPv6 and IPv4 was produced. Some paths paint an interesting picture. I showed the graphs to Bill Owens (NYSERNET) and Joe St. Sauver (Univ. of Oregon); Joe was somewhat disappointed with the picture painted of the path out of Oregon. IPv6 Longterm Comparison I [Matthew Luckie] Did some analysis and graphs as a case study for the IPv6 handout (see the Outreach section of this report). AMP IPv6 Web page [Matthew Luckie] I made some changes to the buildXXX script to accommodate Matthew's IPv6 measurements and did some system manager runs to distribute a new IPv6 measurement host list to the amplets. [Tony McGregor] Spent two days with the team in San Diego early in the month. Bud and I tried the TTL PPS converter which I had sent previously to make use of the HP GPS unit that had been installed by ENS some time ago, with no luck. With Jim, we spent most of the time getting the new pma.nlanr.net server up and going. [Jörg Micheel] The latest release of the Dag6 software is now available and was uploaded to the PMA server to make it available for installation on all new machines. [Bud Hale, Jörg Micheel] The new Dag6 (OC192/10Gig) cards arrived from the Endace Company, and were installed in the Dell 2650 in the machine room here at SDSC. This will prepare the way for data collection to begin. [Jim Hale] While in San Diego, Jörg also met with Chris Gross to discuss the real-time analysis system and what research Chris will be doing down in New Zealand. Chris will be spending the summer (their winter) in New Zealand working closely with Jörg. Before going to NZ, Chris investigated getting stats out of traces. He looked at the NLANR programs that come with the DAG tools, specifically nlanrtsh and nlanrdump. He also looked at some of the code for the Dag programs. After installing the Dag cards on the box that I will be using, I captured some trace files by directly connecting my laptop to the box and doing some IMCP echo requests. I started writing an application that will take data from the Dag card and grab statistics from that record (e.g., average packet size, number of packets). We are planning to move the monitor to the University of Waikato access link (to replace a machine called splash, which previously delivered the NZIX trace data sets), then we will be able to see how the code does under heavy volume. [Chris Gross] ~ Updates on new (and developing) strategically important measurement sites The AMP monitor recently shipped to the AMPATH site at the Fla. International U. is on site, and will be initiated as soon as it is connected (site technicians are preparing the rack space). [Bud Hale] The Brazil machine was recently shipped and is very nearly ready to initiate. However it appears the site in Brazil will be connecting the machine on a different IP and GW than previously provided (and edited into the rc.conf file before shipment). I expect to walk the site people through a single user boot to change it. [Bud Hale] The Great Plains GigaPop site notified us that they are now ready to connect the amp monitor there. [Bud Hale] Received an AMP site confirmation from the Internet2 site in Mexico. I need some additional information but I expect to be shipping this unit shortly. [Bud Hale]
Development and distribution of measurement and analysis tools:Work on the reimplementation (repackaging) of AMP and the new testing architecture continued. I continued to work on the xfer code, can almost send data right through from the ICMP test to the servers. The whole package currently stands at about 7500 lines of code. The xfer code is about 1900 lines of that. I also worked on the amplet data transfer code; there is still quite a lot of work to be done on this, but it is progressing. [Tony McGregor] For St Petersburg, I setup an amp data server on my laptop (under Linux) and transferred the data for amp-hutf to it. The plan is to use this to do a semi-live demonstration. This was useful work because I recorded the process and it will be useful for starting the repackaging. [Tony McGregor] Tried the R statistical language for analyzing the data collected with ipmp_pathchar last period on the WAND emulation network. [Matthew Luckie] Started to add IPMP packet pairs to ipmp_pathchar in order to establish if there were L2 devices present in a hop that was reported with IPMP. Ended up trying something that is a variation on the nettimer technique: sending a single large IPMP packet and then a smaller IPMP that is just large enough to capture all the path information. Assuming no cross traffic (easy to do in a lab environment), the smaller packet will queue behind the larger one on both the forward and reverse paths on every hop - with a few catches - and it is trivial to work out the serialization rate of the path using the separation of the smaller packet from the larger packet at each hop. The initial implementation did not accurately estimate the serialization of each hop. One possibility is that the estimation will be improved by altering the sizes of the large and the small packets. So, I conducted an experiment in the lab where I sent IPMP packet pairs consisting of different packet sizes and tried to make sense of the data with a Gnuplot 3-D plot. I had the size of the second packet on the X axis, the size of the first packet on the Y axis, and the separation of the last byte of the second packet from the last byte of the first on the Z axis. I'm still trying to figure out why gnuplot will not plot a 3-D surface, which would be much easier to view than the dots that it plots by default. I'm sure that I'm passing the data in the correct format. Packet pairs Gnuplot [Matthew Luckie] I found a slight logic bug in Scamper which has been fixed. I worked on the code adding the ability for Scamper to detect if a path is dead after receiving no responses for N hops. Implementation is close to being completed. [Matthew Luckie] Used some data that I got from Bill Manning to obtain a route-view to compare the address list against what is actually routable, and to strip out AS numbers. Ran Scamper over my address list and produced a logical map of the IPv6 Internet (by producing a directed graph and running it through graphviz). The left hand side is mainly Europe, the top-right is mainly North America, and the bottom-right is mainly Asia. 1st IPv6 logical map I gave the same address list and the Scamper source to Bill Owens of NYSERnet and asked if he could run it. He ran it overnight from his Linux workstation, which produces this map when combined with the view from Sorcerer: Owens logical map [Matthew Luckie] Brad Huffaker from Caida rounded out the address list that I had with addresses from missing prefixes identified with routeviews6. Scamper and the address list was run from WIDE, though not to completion. I'm going to contact a few people in Europe and ask if they will run the code. Kenjiro Cho pointed me at a bug in Scamper, which I then fixed, which allowed him to complete his Scamper run on the address list from a machine at WIDE without getting stuck on an address that his machine could not route. The latest logical map: Latest IPv6 logical map [Matthew Luckie] Work continued on the Cichlid reimplementation. Spent time working with Qt/Qthreads API and debugging the code. I wrote new dataset and graph code (not actual opengl rendering, just stubs for now) but hopefully the new interface to the dataset queue management will work well (I redesigned the interface to the Graph class). Rewrote the Cichlid GUI interface in raw C++. I had previously used the RAD developer tool that comes with Qt in order to quickly build a mockup, but my knowledge of Qt has improved enough to let me actually code it. It is much cleaner this way. I also wrote a Qt OpenGL Widget to contain the graph. [Ben Reesman]
Outreach, application support, utilization improvement, and documentation activities:By invitation, Ronn Ritke and Tony McGregor attended the RELARN meeting in St. Petersburg, Russia. Greg Cole (NaukaNet) attended as well. Both Ronn and Tony gave presentations on our work. In addition to giving a presentation, Tony gave a demonstration of AMP (from his laptop, as there was no Internet connectivity). The conference went well and we seem to have built some real bridges with a commitment to host an AMP monitor in Russia and to include measurement in the GLORIAD network. Greg Cole was very pleased with how the conference went. I think it is worth including this quote from Greg in an email he sent to the U.S. participants. [Tony McGregor]
Beginning of the month I spent three days at NANOG28 in Salt Lake City. I used most of the time to talk to folks visiting, but also attended quite a few of the talks. Among the people I met were Simon Leinen from SWITCH, the Internet2 folks from Indianapolis, colleagues from big ISPs, Sprint ATL, UUNET, MCI, and some commercial vendors, Arbor Networks. Had a lengthy talk with Lixia Zhang from UCLA, we talked about joint opportunities in passive measurements and analysis and the possibility of placing a box at UCLA and their ongoing activities. [Jörg Micheel] Attended an afternoon session of the the North American Global IPv6 Summit held in San Diego at San Diego State U (SDSU). Besides the presentations I attended, I enjoyed the opportunity of meeting some of the people attending from locations with AMP and PMA monitors. It is always nice to make personal contact from time to time with people who help in maintaining the NLANR infrastructure. [Bud Hale] Created a handout about our AMP IPv6 activities for distribution at the North American Global IPv6 Summit held in San Diego at SDSU. These were also distributed at an IPv6 seminar hosted by Cal(IT)2 later that week. Ben created a map showing our IPv6 mesh (modelled after his full mesh AMP maps); and Mike delivered them to the proper folks. Bud also helped distribute them at the sessions he attended. AMP IPv6 handout [Matthew Luckie, Maureen Curran] It is hard to judge what the reaction has been to the handout / IPv6 project without actually being there, but amp-memphis, perhaps co-incidentally, asked to be added to the IPv6 mesh this week, which has been done. [Matthew Luckie] A new directory was set up on the AMP server for IPv6. An html version of the IPv6 handout was created to serve as the index page, (using the new AMP page template). AMP IPv6 Web page [Maureen Curran, Matthew Luckie] The latest issue of the Network Analysis Times was also distributed at the North American Global IPv6 Summit held at SDSU. [Maureen Curran] Spoke with Jian-Bo (Univ. Fla) about a future research idea and focus area. [Ronn Ritke] I helped one "gab.seun jones.ewulomi" debug an install of IPMP on a linux system. The system is running linux-2.4.18, so the patch for 2.4.20 did not apply cleanly. I've suggested than an upgrade of the kernel would be easiest. [Matthew Luckie] I helped John Barlow from AARNet get access to a box he had locked himself out of. [Tony McGregor] Responded to a number of data inquiries. [Jörg Micheel]
Ongoing measurement and analysis, networked data, and infrastructure support:Worked on the IPMP IETF Internet Draft to remove the UTC timestamp bit that we added in draft -01, which turned out to be a pointless feature. I forwarded it to Tony and at this time the draft has been through the IETF Internet Drafts editor. Tony posted to the IPPM working group's mailing list. Randy Presuhn submitted comments on the draft. IPMP IETF Internet Draft [Matthew Luckie] Jörg Micheel visited SDSC early in the month. During his visit we conferred on PMA site issues, among other things; Jörg was able to work with Jim to finalize the set-up of the new PMA server. [Bud Hale] A new PMA server was prepared (fitted with an 8 port 3ware parallel ATA Raid Controller with 4 120GB IDE Hitachi/IBM Hard Drives). In the near future we intend to add four more drives to the 4 remaining ports on this machine. [Jim Hale] Site outages continue to be at very low levels in the AMP mesh. Passive monitor sites also continue to operate with few outages. The monitor nai-p-fla (U. of Fla. at Gainesville) is in return shipment (corrupted disk). The site tech asked us to take it back and re-create the system disk. [Bud Hale] Due to a shortage of the Gigabyte mother boards we currently use in the AMP monitors, only three of the four machines ordered have been completed. An inquiry into another suitable motherboard is progressing. [Jim Hale] Monitoring of the AMP and VOLT server data disk fill during the month continued to indicate a rapid fill rate. The fill level of the data disks were already at 88 percent on the AMP server, despite being recently archived. And the VOLT server disk fill level on one disk is at 86 percent. Continued expansion of the AMP meshes has obviously increased the volume of data collected. It appears that data archiving to the HPSS will be necessary nearly every month. I expect to archive both AMP and VOLT in a week or two. I also moved a data directory on the VOLT disk set to achieve an improvement in the disk fill balance on the eight data disks. [Bud Hale] Discussions continued about using solid state hard drives. Jim suggested putting an operating system on a flash memory with an inexpensive USB reader for the purpose of booting remote machines that for one reason or another have become unbootable, or have become unreachable due to configuration problems. [Tony McGregor, Jim Hale, Bud Hale] |
| |||||||
|
Back to the Top last modified: July 28, 2003 
| ||||||||
