Let's replace Ethernet with infrared light bouncing off mirrors! Microsoft-supported boffins hope to eliminate cables in the data centre entirely. They're not, however, deluded enough to think that Wi-Fi is the answer. With thousands of switch ports in a decent data hall, connections have to be uncontested and point-to-point, so the Penn State University researchers have been working on …
The good: distance and figuring out how to lay and order cables becomes irrelevant (for small values of distance). Scaling and multiplexing should work well when there is less hardware (cables) involved. Lasers are cool and are absolutely required if you want to evolve from your 'do no evil' stage.
The bad: Need to stagger units to get the free line of sight. Dust will degrade the connection. The fly or squirrel living under your rack will get in the way of the laser. Laser safety. More dust and light scattering. Solving connection problems involves laser physics instead of some sod following the cable.
Stil it's cool, and you are finally allowed to cross those data streams.
Don't be silly, no need for any staggering - this is where the "and mirrors" part comes up. Preferably scattered along the walls, probably also active, to auto-calibrate them to horizontal (or whatever the plane angle is). You don't even need to arbitrate access, just find one that both ends can see. Hey, I think I'm starting to see the light here...
"Or a single central ceiling mounted reflector hub. Something ball shaped and covered with small flat mirrors that rotates?"
Or planet/moon sized grey metal ball, equipped with a single laser, under the command of Grand Moff Tarkin ??
PS: I still can't believe that thing got into HyperSpace (in Rogue One)
I can't imagine how a university would have missed such obvious problems such as
1. Line of sight.
2. Dust in the air.
3. Animals in the rack.
4. Laser safety
I have to shake my head at their naivety. I suggest you call them up and let them know about these issues otherwise they will spend all that money and get nowhere. You might get a job with them, where you can point out all the things they clearly and obviously have never thought of.
The bad: Need to stagger units to get the free line of sight. Dust will degrade the connection. The fly or squirrel living under your rack will get in the way of the laser. Laser safety. More dust and light scattering.
All of that can be resolved by substantially upping the power of the lasers in use. On first setup they'll make their own holes which also solves alignment, and dust will wink out of existence when crossing the beams, as will squirrels, rats, not-too-awake sysadmins (thus facilitating simple Darwinian selection) and any PHBs who wander into the DC.
In short, from a BOFH perspective this is an excellent idea.
Solving connection problems involves laser physics instead of some sod following the cable.
.. but is a good excuse to expense some truly awesome lasers..
:)
This post has been deleted by its author
Most of the data centres I've been in have the racks so closely packed together, you can't replace old/install new gear without bumping a couple of racks with the trolley as you make your way through them.
So either they're going to have to decrease the rack density, or be prepared for failover-to-cable during replacement runs.
I was thinking about bumping too, but if the software controlled mirror can respond quickly enough (which shouldn't be that hard since it is so tiny) then it won't interrupt the network. At worst there might be a couple millisecond hiccup. If your applications can't tolerate that, then maybe you need to stick with wires.
But then again an awful lot has changed since the 60s.
It's now perfectly viable to have computer-pointed optics at a reasonable price, the lasers themselves are dirt cheap, and the required data rates now are a serious challenge.
Back in the 60s the lasers would cost a bomb, computer controlled optics would be unbelievably expensive. Thus they may not have been trying to use those things, and so would have run into the compromises inherent in using omni-directional light emissions or passive optics. And the data rate they needed would anyway have been fine down a cheap RS232 line.
So I say it's worth revisiting, even if it still doesn't come to anything substantive.
Hmm... I had one of those IR contraptions you could use to hook up your printer to your computer. A pair of boxes you plugged into the centronics port. The IR beam bounced of the ceiling, so line-of-sight wasn't necessary. That was roughly 20 years ago. It did work a such, but never quite as advertised... Ambient light could be a problem. Reflective surfaces in the room could be both a help and a hindrance. And so on. In the end I bought a very long centronics cable.
If anyone is interested: I've still got the things boxed up in my ever mounting pile of obsolete IT stuff; if you pay for the postage they'll be yours... So if your PHB reads about this you can truthfully tell him that you already have the gear to test the idea!
I'm with bazza here - never underestimate the power of convenience-level availability to become qualitative change. What has changed in tech lately is not so much groundbreaking new stuff, but tectonic shifts caused by ubiquitous availability of dirt-cheap high-power tech. Allwinner SoCs were by no means mind-blowing tech-wise - being able to chuck one into anything without even thinking about it was though. Quadcopters don't use any revolutionary new tech - they exist because convenience MEMS accelerometers / gyros / ESCs / LiPos became a thing. Seeing a stepper spin used to involve etching your own PCB full of transistors, buffers and heatsinks - now you just grab a handful of 2x2cm modules out of your "misc Aliexpress junk" bin. We're at the point where chucking the equivalent of a PC into a washing machine controller is the much easier option than making the same thing with sliding contacts, small gears and a motor. I suspect it works much the same with these lasers...
I think these guys are ignoring or seriously underestimating the difficulty of active optics. A design like this will probably be based on piezo actuators if it is to ever be cost effective. But those have a limited range of motion so certain degree of manual aiming is still needed. Now you come into the realm of needing 1000s of these in the same room all intercommunicating, all requiring a certain amount of setup. At what point does it become more cost effective to just use cables?
Also, we have a very good way of getting pulses of light from point A to point B very very accurately. It's called fiberoptics. If anything I imagine the future of data interconnects is in fiberoptic cables (that can be premade and qualified in the factory) which are already far less bulky compared to CATx LAN cables (Especially F/FTP cable).
From the install notes I've read on both CAT5(e), CAT6, CAT7 and various fiber optics products the minimum bend radius was something like 150-250 (CAT) versus 200-300 mm (Fibre). Not exactly an earth shattering amount of difference, and certainly enough to accomodate easily within typical equipment racks, ceiling ducts and raised flooring. And are we talking U/UTP or F/FTP? (Or any variation between?)
There is also a large difference between the installation minimum radius (Bend it once and leave it there, which is tighter for copper) and the dynamic minimum bend radius (Keep bending it, in a frequently used patch rack or a cable slab on a moving piece of machinery, in which fibre frequently has a smaller bend radius than many copper options)
Copper's minimum bend radius guideline is about two things. One, attenuation, because RF doesn't like making turns, but attenuation only matters for longish runs. If you have a 15 meter run instead of 100 meters, you can survive a lot more attenuation so it will work fine even if you've made several loops around your finger. Two, mechanical stress, i.e. solid copper conductors will break if bent and re-bent too often. That's not a problem the first time you bend them, but if you do it often, especially in the same place, the conductor will eventually break. That's why you use stranded copper for patch cables, since they are more likely to be moved around a bunch.
You can cheat the bend radius to a FAR greater extent in copper than you can with fibre.
"It's also a nice practical demonstration of why images taken using these cameras don't match what our eyes see."
An older one was photographing some blue flowers with Kodachrome and Ektachrome (which I'm glad to see they're bringing back). Kodachrome would render them pink, having detected the reflected IR.
Is this the beam in the recent film where it plainly goes FTL in order to jump between star systems which are quite a lot of Light Years apart but somehow is considerably slower when it nears the planet allowing the terrified inhabitants ample time to scream? Quite apart from the joke of multiple energy beams converging and changing direction and then doing the reverse on arrival (while being perfectly targetted at each planet).
Yeah, one can think on these things too much :)
I just see this as trading one set of problems for another. Imagine an technician pulling out one system to fix or replace it and physically blocking a bunch of IR signals in the process.I have an IR remote on my TV and dread the problems you will encounter based on that experience.
The elephant in the room: The problem is the connection speed between computers. We are limited by modulation speed, not carrier frequency.If you can toggle a bit at 1 Gb/s, you want a carrier frequency of 8 GHz or higher, 10 Gb/s = 80GHz or higher. Moving to THz and higher frequencies does not increase the bandwidth between computers. Physical connections (as opposed to wireless) between computers prevents cross-talk, interference, and inadvertent blocking. This reduces the amount of bandwidth needed do to lost bits you are very likely to get through a wireless transmission system.
My recommendation, stick to wires (or optical fiber). Look at materials that have higher switching speeds than silicon, such as GaAs, InP, etc.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
