FTTH Trial Status
Monday, October 22, 2001
Based on a Sun, 21 Oct 2001 email from Peter Allen:
So far almost everything we have seen thus far while lighting up our Trial network is common to that of any LAN starting up. There does seem to be inconistencies over VLAN and Internet site accessability related to MTU (Maximum Transfer Unit) settings for the WAN. Meanwhile, my wife is thrilled with the bandwidth to her iMac. Other folks in the Trial are fine with the FTTH connection, but in trouble over understanding and managing their own home networks. For this we have house calls from the 'nerds' and a learning experience for everyone. One of the things that we have learned is that we all have great neighbors. :)
Based on a Tue, 16 Oct 2001 PA-FiberNet post by Marvin Lee:
Several people have asked for a discussion meeting to help bring them up to speed with their new FTTH installation and to open questions about our new found bandwidth capabilities. We have arranged to use the kitchen/meeting room in the Art Center, Thursday Oct. 25, 7:10pm. Enter the usual entrance and we will be down the hall from our usual room.
Wednesday, October 10, 2001
Links of Interest, Reminders, and Recommendations:
Don't forget to stay abreast of the broader view on fiber connectivity for Palo Alto citizens at PA FIberNet's web-site: http://www.pafiber.net. In particular, I recommend their news index. Hilda Weisberg is doing a great job as web-master for that site.
David Rosenthal gets credit for the following pointer to a Northern Sweden FTTH project: http://www.acc.umu.se/~tfytbk/mattgrand/.
Anyone with a broadband connection has a responsibility to help prevent distributed denial of service (DDoS) attacks and needs to understand this threat. Back in June, Cindy Bickerstaff posted a pointer on PA FiberNet to an excellent site, which is worth repeating now that the trial is up and running. See: http://grc.com/dos/intro.htm.
grc.com also has a review of software firewalls at: http://grc.com/lt/scoreboard.htm. Note in particular that grc.com gives poor marks to the BlackICE Defender software, which I had discussed elsewhere on this site. For a number of reasons, I'm currently using a hardware firewall product similar to the Linksys product mentioned previously on this page.
Based on a Wed, 10 Oct 2001 email from Peter Allen:
[D. Rosenthal's 9 October post is a good one. The trial needs this kind of scrutiny. On a related note, last] night about 11:30 it took me about 1/2 minute to download an 8MB file from Microsoft. This is about a (64Mb/32seconds=) 2Mbps rate. I did this twice. While the speed was really nice, it is still half of what I expected at such a late (and hopefully quiet) hour.
Tuesday, October 09, 2001
Based on a Tue, 9 Oct 2001 email from David Rosenthal:
[I want to expand upon and clarify the performance problems that I first described in my October 2nd note.] People I spoke with were complaining that their FTTH service was *much* slower than their cable modem or DSL service, so slow that they had given up and weren't using it. I do not know exactly what was causing this performance, but I have no reason to believe their reports were false.
If the trial is to succeed, it is important that the participants believe that they are getting good performance for the price. That's not what the people I spoke with were reporting. [Expectations of FTTH performance were set at very high levels as the technology the trial initially planned to use was incredibly fast. In contrast, while the Marconi technology the trial ended up using is fast, it is not earthshaking. In addition, cost limitations significantly limit the trial's aggregate bandwidth upstream of the head-end.] Thus it is very important that people actually get the potential performance of the Marconi technology.
My observations show that given correct configuration of the Marconi box, the Linksys router, and the network on the home side of the router it is possible to get close to the rated performance of the FTTH link. The route I have been measuring was carefully chosen - it goes from the head-end via a via.net router to PAIX and then via Stanford's PAIX gateway to one of the Stanford core routers. This is a rather short, very stable route with no bottlenecks other than the FTTH link. Installing a server at the head-end is not likely to show much difference. I did in fact measure performance to a server at via.net - it was awful. Much worse than to the Stanford network and only barely better than dial-up speed. It is not clear what caused this.
It will take some time to get everyone's configuration correct. My intent in posting my measurements was to encourage people who aren't seeing good performance to work with the Heyertech support people to investigate why. Unless they have a complete picture of how the system is performing it'll be hard for them to fix the problems. You *can* get consistent 3.5-4Mb/s transfer rates. If you aren't, tell them.
Based on a Fri, 5 Oct 2001 email from Ron Krasnow:
Update on my non-working connection ... the card and splice in my Marconi box was replaced and fixed, respectively, on Wednesday, but the connection to my home still does not work. Since the other 3 homes on my "node" are up and running, the thought by Heyertech is that the "splitter" in the street is not working properly. I was told that the city would fix that, but no date or time has been set as far as I know.
Wednesday, October 03, 2001
Based on a Tue, 3 Oct 2001 email from Peter Allen:
I myself was out of service for 10 days (once my Marconi box was installed). The problem was with the blade (card) in the Marconi equipment at the head end (the 1310 nm tranceiver was inoperative). While this was a first test of both our diagnostic as well as responsive capabilities, we should expect to learn things from a Trial, right? We have. Once this problem was solved my connection came up - so flawlessly in fact that our (English and technology challenged) Au Pair was able to make it work (in minutes) while I was on the phone with him.
As for Trial folks bummed that they can't get more than 3-4 Mbps on their downloads, this seems laughable to those folks still constrained by DSL, Cable, or (ACK!) Dialup. This is like a Porsche driver complaining that our highway are only designed for 90 MPH traffic. Yet we need to test this system - and we need to 'break' it by showing that there is demand for speeds of more than 10 Mbps. So I'm afraid that I will continue to encourage the whiners. :)
Meanwhile we should alert folks that the bandwidth that one will get from some of these external test URLs and downloads will a) not likely approach 4.5 Mbps (our worse case speed) and b) will vary each time. This is due to a) other upstream bottlenecks and b) different trace routes used each time. This is similar to three letters bound for NYC mailed on the same day, yet arriving at different times. I suggest that we get a box to stream ping at the PAIX and one within our network. I volunteer to host a Linux box for this purpose (and other tricks like Hollywood squares videoconferencing).
Based on a Tue, 3 Oct 2001 email from Ron Krasnow:
[. . . ] FTTH is not working for everyone. I've had a box for 10 days and it has never worked. There are Marconi box problems, splice problems and other problems that have not been spelled out for me (something about the head end). [. . . ] There are workers here today supposedly fixing things, but we'll see....
Based on a Tue, 2 Oct 2001 email from David Rosenthal:
[People have expressed disapointment to me regarding the performance of their FTTH connections.] I've made some measurements. FTP downloads of large files from machines at stanford.edu and mozilla.org consistently run between 3-4M bits/sec. The configuration of the network on the home side of the Marconi box can make a significant difference. Using switches I was seeing 3.6-4.2M bits/sec, using hubs I was seeing 2.8-3.4M bits/sec. These measurements were taken using a Linux client. A Windows client gave roughly the same results, but I don't have accurate measurement tools for Windows.
I did see much slower performance talking to some other sites. I am in touch with Heyertech to look into a possible misconfiguration upstream of the head-end.
The performance measurement site at MSN/Cnet consistently gave highly optimistic and seriously misleading numbers, and should not be used as a basis for comparison.