Slashdot Mirror
Good Demo System For A High-Bandwidth Link?
FuzzyDaddy writes "My company is planning on demonstrating a 2.5 Gigabit per second link to some potential customers in the next few months. Now, we have all the equipment needed to measure how well the link is performing, but we'd like to put together a cheap 'Gee Whiz' demonstration. Surely other /.'ers have put together similar demos in the past. What combination of computers, network adapters, and software have you used to demonstrate high data rate links to potential customers?"
See how many DVDs you can have going @ the same time. Copy some DVD files to a hd (smartripper will do this), and share them over the link. Use a software player like Media Player Classic if on win, or your fav player on linux. Load up multiple instances and see how many of the movies you can stream.
Also you can repeaditly stream a text version of War and Peace or some other lengthy book, with a counter on the recieving end showing how many times you have downloaded it. Keep a copy of the print edition on the table to show them what is comming down as the counter ticks away.
------- Assumption is the mother of all f$#@ ups.
People might like to see how their data is safequarded, you could do a complete backup of 1 terabyte of data in under 7 minutes. That might be, "like WOW." Give a 7 minute presentation during the backup..
i work at CITI and we've had recently done a few demos with our high-bandwidth link. one setup included two dell dual-CPU servers, one at either end of a gigabit link. we then used iperf to fill the majority of the link with traffic (using other machines). we then used a CITI project with the intervening Summit 7i switches to reserve bandwidth for a video teleconference. we demonstrated the practical capacity of the link and the ability to honor QoS parameters.
the CITI project used to manage the switches is, among other things, a secure remote invocation architecture that we use for a related network testing and performance-oriented umbrella project. that project's ultimate goal is to provide a distributed, real-time router-to-router traffic analysis system for use in optimizing campus networks and isolating networking failures. check our the web page if it's of interest.
d
.
Do a 'time machine' demonstration. Throttle the bandwidth to, say, 56k and explain that this was 'The Internet' ten or twelve years ago. Demonstrate some moderately taxing application for the time (like a large download).
Take the audience forward in time by increasing the bandwidth slightly. Note how the previous application just zips by now, but start a new application that's still slow.
Repeat a few times going through a sequence something like: download large file, surf web, audio, tiny little image of fuzzy movie, voice-over-ip, real-time video with crappy quality, real-time high-quality video.
End the presentation with a question mark: every new level of bandwidth made previous uses easy, and enabled new applications that really needed the bandwidth. What will be the new application that makes you glad you have 2.5Gb?
It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail. - Abraham Maslow
If it is to the Internet, then you're out of luck. 99% of anywhere you go couldn't come close to filling it.
Between remote offices? That is much better. It allows for things like multi-camera video conferencing or multiple simultaneous conferencing sessions.
It also lends itself to "location transparency" demos -- where it doesn't matter where in your system the resouce is, it acts like it is right at your fingertips.
For example, realtime video/audio editing of multiple tracks while the raw data is stored on a SAN in one building and the editing horsepower is in a different building -- and you're in a third just piping the interface.
Large scale CAD/Design reviews, with people being able to mark up and manipulate 3D imagery in real time, regardless of where on your net they're at.
Your big problem is going to be device latency. Spinning drives up, delays in software starting, etc. is going to be much more noticable. Bandwidth like that is great once the bits start flowing, but getting it started and keeping it filled will be taxing.
Unless you do testing that generates obscene amounts of data -- like collecting data from a supercollider, etc.
-Charles Hill
Learning HOW to think is more important than learning WHAT to think.
You could set up an access-grid node:
http://www.accessgrid.org/
It's got a pretty-good 'wow'-factor, and is one of those things that people instantly want at their own site. The coolest thing is that it scales; it runs on hardware ranging from a laptop with a webcam to a custom-build facility.
The comfort you demanded is now mandatory - Jello Biafra
Nice regurgitation of half digested technical articles. 3GIO *IS* PCI-express, its just the old name. Oh, and most PCI-X interface controllers have multiple DMA channels allowing you to send data in either direction. In this case a single PCI-X part at just over 500MB/s can more than saturate a 2.5Gb/s link so the fact that the bus is uni-directional during any single transfer is moot. The individual DMA channels have separate buffering allowing very clean and seamless DMA transfers at FULL speed. And BTW, most of this fancy PCI express tecnology you mention *doesn't exist* at the practical level today. The bridge spec was just finished in June of last year and the first compliance workshop was last december and this was for *baseline* PCI express implementations.
The bus has nothing to do with it, its the card and how it (Drivers primarily but also bus interface, buffering, etc) has been implemented. We transfer 300+MB/s sustained across shared PCI-X 100Mhz slots with less than 10% CPU load, the source is a second (133Mhz) PCI-X card that is feeding the two shared slots.
Make sure that the system you are using has enough PCI-X channels. Most motherboards that have SCSI/SATA down on the motherboard tend to share the PCI bus with the wrong slots. For this reason we run a 3 PCI-X slot motherboard which has minimal integration on the motherboard. This allows us to use one PCI-X 133 slot for the dual U320 SCSI controller (SATA isn't ready for this yet) and the other two slots to run multiple high definition channels (uncompressed) at 4:4:4 sampling. Thats quite a bit more bandwidth than we are talking about here. The system runs so clean that I have played Q3 in the foreground while all of this is going on. Oh, and this was all happening on Windows XP Pro.
I would also suggest that whatever you try to send, use UDP not TCP/IP.
Its not the motherboard, its not the OS, its not the slot. Its the engineering and not the kind that comes out of reading press releases.
I think the important thing here isn't the 2.5gbit link itself, but how it performs in comparison to other highspeed connections.
If you have one computer (the one on the 2.5gbit link) streaming 300 DVDs in realtime, and another computer streaming 1 with an exremely jerky and possibly laggy DSL/cable connection, it will allow people to grasp the depth of the situation. Just showing that it's blindingly fast won't do anything for you if you can't provide a baseline from which the average joe can compare.
Somebody else suggested having it download "war and peace" over and over, while having a hard-copy sitting nearby so you could have something tangible to say "All of this information is being transferred from office to office in a matter of seconds. With this kind of highspeed link, e-mails with attachments the size of the statue of liberty would be received almost instantly." etc.
Geeks know what 2.5gbps gets you. Real people don't, and you need some way to contrast the power of their current internet connection with the power of the new proposed one.
Doing multiple things at once, if not the playing of multiple DVDs, is what's going to win people over. Streaming video gets the layperson response of "My TV does that" (as another commenter pointed out). However, if you can have a computer displaying every single cable channel available all at once or something along those lines, then THAT would be impressive to the average joe. Or perhaps a video conference with hundreds of remote parties? Each client connection would have its own bi-directional video stream, such that the clients could see the person doing the presentation, and the person presenting can tile 100 windows on a 2048x1280 (or whatever high resolution) screen, all showing a different person in a different place all in fluid motion, in realtime.
Reinvent the wheel only at either a lower cost, greater effectiveness, or your own personal enrichment and satisfaction.