Slashdot Mirror
Is a Wireless Data Center Possible?
Nerval's Lobster writes "A team of researchers from Microsoft and Cornell University has concluded that, in some cases, a totally wireless data center makes logistical sense. In a new paper, a team of researchers from Cornell and Microsoft concluded that a data-center operator could replace hundreds of feet of cable with 60-GHz wireless connections—assuming that the servers themselves are redesigned in cylindrical racks, shaped like prisms, with blade servers addressing both intra- and inter-rack connections. The so-called 'Cayley' data centers, so named because of the network connectivity subgraphs are modeled using Cayley graphs, could be cheaper than traditional wired data centers if the cost of a 60-GHz transceiver drops under $90 apiece, and would likely consume about one-tenth to one-twelfth the power of a wired data center."
Unless they plan to use microwave beams for power.
"Beware of he who would deny you access to information, for in his heart he dreams himself your master."
Or Rlly? So a traditional datacenter is sinking > 90% of its power into the wired network connections? Not the actual servers themselves? Not the cooling? The wired network connections? I'm not buying those power saving estimates.
AntiFA: An abbreviation for Anti First Amendment.
Can someone explain how a wireless approach could use less power than a wired approach?
I understand that if you compare a crappy wired implementation to highly optimized wireless implementation the wireless might win out,
but then it would be cheaper to optimize the wired one.
until the wackadoodles who claim they get headaches from radio signals find out they're living next to a place which runs such an environment.
I can't wait to see the signs they use to protest as they stand outside in the blazing sun:
Stop killing us with radio waves!
Radio waves kill!
Save a life. Turn off your radio.
We will bankrupt ourselves in the vain search for absolute security. -- Dwight D. Eisenhower
They obviously mean computing power. Because the networking is so bad that all processes are blocked on I/O most of the time. :-)
The Tao of math: The numbers you can count are not the real numbers.
You can't have nearly infinite bandwidth in a finite frequency spectrum, but you can keep adding a shitload of wires if needed.
Given the problems people have when multiple wi-fi routers are too close together like in an apartment building, I am doubtful that it would work well in a server environment, not matter which frequencies are used.
When the 60Ghz transceiver (which doesn't exist yet commercially) drops to $90 each, won't 10Gig ethernet drop down to $9/port, skewing their cost justifiication results? They mention using 4 - 15gbit transceivers... what's the aggregate bandwidth of a 60Ghz network? If the aggregate bandwidth is 15gbit, that's not going to handle a rack full of servers.
60 mHz means about 1 cycle every 16.7 seconds.
The Tao of math: The numbers you can count are not the real numbers.
the traffic is sent into the air and its up to each receiver to filter the noise and ignore data not meant for it. lots of interference.
its OK for starbucks or for home use but not by much. i have at least 10 wifi networks around me that constantly interfere with mine. i used to get regular disconnects from x-box live that went away when i tried to connect my x-box to my router with Cat5 cable. same with video streaming.
this is why large events have crappy data speeds. everyone is broadcasting into the same air space and interfering with each other.
I am so happy that Microsoft is doing that kind of loony shit.
Contrary to the popular belief, there indeed is no God.
Indeed, you have to do the opposite: Turn all radios on all the time, so that they can consume all those evil radio waves before they reach you. :-)
The Tao of math: The numbers you can count are not the real numbers.
So Slashdot is now ripping off other sites, copying their content to Slashdot-hosted pages, adding ads, and breaking links. The original article says "Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ANCSâ(TM)12, October 29â"30, 2012, Austin, Texas, USA. Copyright 2012 ACM 978-1-4503-1685-9/12/10 ...$15.00."
In the actual paper, the power consumption bullshit part reads "Power consumption: The maximum power consumption of a 60GHz transceiver is less than 0.3 watts [43]. If all 20K transceivers on 10K servers are operating at their peak power, the collective power consumption becomes 6 kilowatts. TOR, AS, and a subunit of CS typically consume 176 watts, 350 watts, and 611 watts, respectively [9â"11]. In total, wired switches typically consumes 58 kilowatts to 72 kilowatts depending on the oversubscription rate for datacenter with 10K servers. Thus, a Cayley datacenter can consume less than 1/12 to 1/10 of power to switch packets compared to a CDC. That's comparing transceiver drive power with a whole store and forward switching fabric.
It's also not clear how their "Y-switch" thing, which doesn't store anything, handles busy reception points. At some point, in a forwarding network, you either have to store packets or drop them. Or set up end to end channels first.
The More You Know.... *comet*
Hopefully they will also pass out those cancer detecting bras to all of the staff members as well.
The overall amount of radiating energy involved would make a datacenter technician ... medium well.
Sent as ripples into the electromagnetic field. No single photon has been harmed in the process.
Even with careful planning and management, wouldn't a completely wirelessly-networked datacenter be more of a target to hacking? Even with a high level of encryption, which would add to network overhead?
Are YOU using the TOOL, or is the TOOL using YOU? Think about it!
Cheap receivers radiate more than the signal they are receiving. How else do you think radar detector detectors work?
Learn to love Alaska
Will be the solution. If only we dared to wait for the technology.
Sent as ripples into the electromagnetic field. No single photon has been harmed in the process.
so-called: (adj) alleged, supposed (doubtful or suspect) "these so-called experts are no help"
I'm getting cancer faster just thinking about it.
I wonder how robustly Microsoft plans to address security at a wireless data center. In many data centers, wireless devices, even encrypted ones, were simply forbidden and twisted pair was inside physically locked metal conduit. Most security schemes for wireless transmission will involve more overhead on CPU, memory, transmission and therefore, energy, air conditioning, floor space, etc., not to mention a staff division related to spectrum monitoring & analysis.
On the other hand, if the data center is merely for storing consumer's account information...[ rimshot ]
...Is this real life?
60GHZ would barely get through a wet tissue. You could track the location of the technicians by watching the server-down warnings move around.
No. "Team" means a group of individuals working together. Not more, not less. They usually do so to achieve a goal, but that's not part of the definition. And it's definitely no given that a single member of the team couldn't achieve the goal.
The Tao of math: The numbers you can count are not the real numbers.
It'll be cool when somebody microwaves a burrito in the lunch room and random servers drop connection. 3.. 2.. 1.. ding! Hm, server connections are back.
Modest doubt is called the beacon of the wise - William Shakespeare
I was always a bit dubious of the infrared based wireless networking (like IrDA) for an office environment, but what about optical wireless in a data center? Seems like that would solve the potential security issues and you could isolate racks (or parts of racks) on their own wireless network and then do the traditional wired scheme to join those nodes together so that you weren't stretching the bandwidth too thin?
A grad student at UCSB recently gave this presentation: http://www.cs.ucsb.edu/~xiazhou/papers/sigcomm12_beam3d.pdf
These days, with VMs (and hence software switches) carrying the actual workload, and hastily programmed core switches broken down into a hundred VLANs, why are we hanging on to the ancient notion of "wires"? Clearly a wireless method for every server to be able to talk to every other server is the next logical evolution. Just sprinkle a little software on top to make sure that the servers only see/process what they are supposed to, and surely it will all work great!
As someone who makes a living installing and supporting wireless data communications, and has done so since 97, this is totally laughable. Not to mention insecure as hell. And did I mention unpredictable performance? And how did they measure latency on a draft spec that is still a horse race between 4 draft specs? Looks like someone has a new buzzword for when "cloud" starts to wain.
Within a data center, you could use $1.00 LED emitters and receivers with integral lenses for short runs, precision (but still cheap) alignment fixtures and $0.10 mirrors. For long runs, LED laser emitters. You'd still beat $90/point by a huge margin. And as a plus, you'd have some extremely high speed connections. Power consumption... I dunno, you'd have to do an analysis. One thing that seems obvious is that for any line not sending data, the LED should be off the vast majority of the time.
I've fallen off your lawn, and I can't get up.
Hey, I have this awesome idea, let's take out all those expensive copper wires and make our data center wireless. It'll save so much money! But first we'll have to redesign racks to be cylindrical and servers will need to be keystone-shaped. Also, because of the new rack design, you won't have access to rear ports. If something in the center of the rack comes undone or stops working, you need to open the entire rack. And each rack will have to be a faraday cage so the signal doesn't leak out and collide with other racks. And each rack has to share the bandwidth. And servers will lose their connection every time a mouse farts. And you'll also have to devise a new cooling method because hot/cold-aisle won't work with a bunch of cylinders.
Just think of the simplicity and savings!
Yeah, I had a bit of trouble with that as well.
Here's why.
I have a pair of 4U servers, each containing a 1000W PSU (for hard drives and fans) and a 450W (for the mainboard and everything else). That's shy of 3kW on servers which wouldn't fill a rack a quarter the way. The switch (24-port unmanaged) consumes 40W. That's about 1% of the total power requirement of the entire system. If I switched out for say a Linksys E3000 (7W) and the associated wireless interface cards (I'd have to go with USB since all my expansion slots are occupied with SATA controllers, so call it 5W a pop), I would save a staggering 23W (nowhere even close to 90% of the total power requirement of the system) and all of nine feet of copper, on top of creating a data security nightmare.
Operation Guillotine is in effect.
Problem - 60GHz is currently very near-space wifi. It's also what, a couple of gigabit worth of bandwidth. Also, I haven't seen any studies yet looking at 60GHz saturation and lots of multi-path reflection. It's a cool technology but it does read like someone's trying to sell the tech, rather than really being suitable for it.
Not really, no. Want two nodes connected in your network topology, just pull a wire or fibre between them. Requires lots of wires, but it is going to be better than wireless on throughput, latency, and reliability. Wireless is useful for devices that are constantly moved around. For anything stationary, pulling a wire is a better long term solution. And I assume servers in most data centers are considered stationary.
I know people who'd love to see wireless data centers become a reality (assuming they don't have to debug new problems). But I consider it to be a pipe dream. Wireless will never catch up with wired networks, because any wireless technology could be turned into something more reliable by using some sort of wires to ensure the signal gets where you want it, and not everywhere else.
Do you care about the security of your wireless mouse?
You don't understand how it works. Your radio causes vibrations and oscillations in the same magnetic harmonic frequencies as the transmission. These vibrations upset the natural rest state oscillation harmonics present in All Living Things, these negative and deathly vibrations cause cancer. Life Crystals oscillating in the same Resonant Frequencies absorb these energies and give off life-giving restorative vibrations.
A radar detector works by detecting the radar sent by the radar gun. In some states (VA?), such detectors are illegal, and they use detector detectors to find users of them. They work by detecting leaky receivers. I think you misread my comment.
Learn to love Alaska
This will only work if the data-center is deployed as a PaaS cloud grade apps. Falling to leverage those best of industry game changing paradigm the interference from all that vapor will have a detrimental effect on the TCO and ROI KPIs.
Jehovah be praised, Oracle was not selected
What you say is, true, despite the humor. Let's say there's nice low power low noise wireless. Multiple concurrent channels would have to wired in a method that's like a cross-bar L2/L3 switch to make this work, and the matrix (sorry for the word choice) would have to service each point with non-blocking architecture as the bottleneck built in cache could be enormous and fast. The engineering costs of this outweight the perceived savings. Add-in any random EMI (admittedly at really high freqs), and this somebody's pipe dream.
---- Teach Peace. It's Cheaper Than War.
From TFA: "the authors picked a Georgia Tech design with bandwidth of between 4-15Gbps and and effective range of less than or equal to 10 meters."
Provided interference, does that mean you won't get more than 15Gbit per second for all the machine in a circle of 10 meters? How much is that 6 racks? You put what, 20 machine per rack? (I am not in IT, so I am not exactly sure.) so you share 15Gbit per second accros 120 machines?
Assuming no other interference. Right now, you can get 10Gbit per second with 10gig ethernet full duplex from all the odd machine to their even one. that gies you a cross cut of about 600Gbit per second.
Of course, you do not have that much on the inside. But if you are running an hadoop cluster (for instance), you definitively need that bandwidth!
I really don't understand in which kind of server room, you would want wireless networking.
Since wireless is not all that difficult to break into, I'm not surprised to see the Microsoft name in this article. What price do you put on security? Is it worth giving your corporate information to anybody nearby just to save a few dollars?
That 40W switch does things like VLAN, QoS, etc that I rather imagine the Linksys would choke on. that 40W switch probably also handles 48-96gbit/s of traffic, which I KNOW that linksys cant handle.
Bigger problem: Im not aware of a "non-blocking" wireless AP.
that's the thing - it doesn't. It's *unmanaged*. It's a D-Link DES-1024R+. Just a dumb switch. It doesn't even do DHCP. As to saturation, it's a nonblocking wire-speed architecture around 4.8Gbps via 24+1 autosensing ports (the +1 is a bay for 100MBit optical which is in another switch (that one a D-Link 16+2 with two optical bays but I can't remember the model number). Oh yeah, and it's noisy. I think the fan bearing's gone.
Operation Guillotine is in effect.
Somehow, they're concluding that 90% of the power used in a datacenter is used for network adapters, switches, and routers? Something smells rather funny here...
I have no problem with your religion until you decide it's reason to deprive others of the truth.
What makes you think it consumes 40W? Have you used a power meter on it? Same goes for your "1000W" servers. Power supply ratings do not indicate typical consumption levels, only peak load capabilities, and the only time you get even close to that is when you're first powering the system on and spinning up your drives.
Because I've metered it? And yes I am perfectly well aware of the peak load on systems when you spin up the drives, which is why there are TWO power bricks in each server. The specifications on each are identical. The peak load on the kW bricks on spinup is 860W, steady and stable(ish) at 412W. The system bricks peak at 383W and stabilise at 241W (saturated), 191W (idle).
Operation Guillotine is in effect.
Ok, those are very realistic numbers for a server.
Simply put, in order to pull that off, you'll need fairly sophisticated data processing, simply pointing 2 directional antennas at each other works fine outside, but is much more problematic in a data center which has walls and obstacles creating reflections. So what you could do is to use modern MIMO systems, but that would require huge amounts of processing power to get any kind of decent bandwidth. It's no point designing a system now which already peaks out at 10 GBit.
Just because you have a particularly weak and inefficient switch doesnt mean thats par for course. 40w would be about right for a high-end 3000-series Cisco which WOULD do all that fancy stuff (except generally DHCP is NOT something you want your switch doing, since it is distinctly Layer 3; not sure if even 3000-series switches do dhcp).
And for the record, if your switch is non-blocking, its doing 9.6gbit/second of traffic at 24 ports and 100mbit port speed (full duplex=200mbit/port *24 ports). Good luck at getting a tiny fraction of that speed out of a wireless access point. Switches tend to scale really well with multiple hosts, and dont have to worry about collisions; WAPs tend to scale incredibly badly, partly because they do have to worry about collisions.
Too bad the people who work there won't be allowed to use their cell phones during an outage. They'll be running back and forth to their IP phone relaying messages. Its bad enough that many data centers officially do not allow cell phone usage but everyone does it anyways because its just not feasible not to use your cell phone. But in a wireless data center it would be strictly enforced. It would be a nightmare trying to get work done there.
I guess this would mean that one would require breathing gear and an oxygen free atmosphere in these type of data centres to get the optimum wireless performance. In the 60Ghz ISM band where Oxygen reduces the transmission due to absorbtion maybe these super green centres would become O2 free environments with operators doning breathing apperatus in the data centre to get the best performance out of the wireless transmissions ? -- patent time !
bæ8Ã0sÃOE?5r©oÂÃ?âz:ÃÃAÃ?ÃOEÂ6fXÃ?]Â
It looks like their hope with the cylindrical orientation is that each server will communicate directly with the 5 to 7 servers opposite it via the inside (and hopefully the signal would be absorbed there as well) and with the servers above/below it on the outside (where the signal would dissipate fast enough to not interfere with other cylinders). Quite intriguing, but it creates one giant (and complex) software-managed ether in the literal sense, information will just "be there" and hopefully the software will keep it moving in the right direction. Good case for IPsec in the data center; hopefully the blades they design will have good hardware to handle encryption otherwise this model will only be good for single use "supercomputer" type applications (where cabling is a minor issue anyway).
That's like saying, "with all these fancy new technologies in hybrid cars with GPS and everything, why are we still using wheels?"
Wires are cheap and reliable, and even if they get eclipse in bandwidth at some point by wireless technologies, they will have a place for a long, long time for those reasons.
doesn't make it a good idea.