People are using cell phones and killing their landlines. We have wireless networks in the home to connect our myriad devices. And maybe wireless is coming to the data center, as well. In days of old, servers kept to themselves pretty much in the data center, and interacted with end users out there on the Internet and the …
I wish the author had of used ToR as the acronym, rather than TOR. I started thinking about The Onion Router project rather than Top of Rack switches and my little brain got all befuddled.
I wonder how the security of these backup wireless backbones performs? If companies are able to get 'boundary' protection wrong, transmitting back-end data centre traffic through the air fills me with dread that someone will find a way to listen in and have their dirty way with my data.
Wiring (fibre or copper) outperforms Wireless.
Either the reporting is bad or the people doing this are clueless.
Or perhaps you didn't read the article properly.
This is primarily aimed at at situations where the physical network is saturated both in terms of bandwidth and physical connections. If the situation is temporary or short term why on earth would you spend huge amounts of effort reconfiguring several racks when you can create temporary wireless connections?
Physical is faster but it's not necessarily always convenient - that's the scenario this work is targeting.
In my experience "temporary" solutions quickly become permanent installations.
I would much rather drop in a bundle of fibre for £1 per meter and put an interface on it. Only a fool puts a single pair in to a rack when it is relatively cheap to put a bundle in, if you run out of capacity between racks you can just terminate another pair. The biggest risk is the top of rack switch being saturated, but adding a wireless unit to the top is probably more expensive than adding another wired switch. Adding a 60GHz modem has to cost as much as a optical interface. You can even do a server to server link over optical, or link the server to another switch with optical.
Data centers to cut LAN cord? NO
Wireless has a finite spectrum and as such places a limit upon the numebr of servers talking at once before you start impacting each other. Wires do no. WIreless also add's another security level you need to cover and then it's enough for most data centers to balance there 3-phase power usage, let alone wireless spectrum used in which server in what rack.
DC power and data of power lines would be more usable than this, maybe for monitoring of servers, then this has possibilities, but as main networking pipes in and out of the servers, nooo.
We used 10GbE for "east and west" traffic. It outperforms WiGi and is limited to the confines of our cables. i.e. no chance of wireless hacks. 10GbE has a higher initial cost but it's seriously fast and means you never have to worry about internal traffic saturation for the life of your architecture. (We redesign the architecture and replace everything every 3-4 years.) In fact we bonded 2x 10GbE together to give up to 20GbE of bandwidth.
I'm a bit puzzled. Yes, you can make wireless links work in the datacenter by using what is basically a P2P approach. However, wired links are (and, as far as I can see, will always be) faster and more robust than wireless. In a datacenter cabling is easy; you could set up the same "P2P" network with cable and get 10x the speed. Without having to worry about the next solar flare or the guys in the unmarked van across the street (the ones with the cheesy slavic accent. Or slanting eyes, long fingernails and a long sparse beard. Or all of the above). Wireless is a very good compromise for some things (instant setup for new kit, you can move your kit around, no need to buy expensive cable, the missus won't bitch as much about the cables all over the place etc.). But in a million-dollar, fixed, dedicated structure like a datacenter... I only have a word: why?
(honest question, btw)
I am thinking it is mainly for the same reason...
...non technical people don't understand that 'broadband' describes the form of the transmission rather than the speed and don't understand that, no, their wireless 'broadband' will not go anywhere near as fast as the 'broadband' that comes down a copper wire.
IE: there is always a sucker (PHB) with a chequebook around.
They also get really confused
If you start talking about 'narrowband'.
The term broadband has become a bit of a cover-all of late though. I was contacted by someone for help the other day who said he was 'on dial-up broadband', which was very clever I must say!
The issue with laying new cables for east-west is that it can quickly become expensive. Switches aren't particularly cheap and cabling can be quite expensive, so this idea plugs a hole from what I can see.
That said, IMHO it's a very beancounter way of doing things;
At the moment, saturation doesn't happen too frequently and therefore the Wifi route is the more cost effective route (why install physical cables for something that will rarely be needed?). But if the pundits are to be believed, and everything will be in the Cloud (well except my data!) then there will be a lot more east-west traffic. At that point new cabling may be the better option.
Of course the PHB will be after what's cheapest now with very little thought of possible expansion and future scenarios.
As the paper says:
"In prior work, we argued ... for a more modest addition of links to relieve hotspots and boost application performance. The links, called flyways, add extra capacity to the base network
to alleviate hotspots. When the traffic matrix is sparse (i.e. only a few ToR switches are hot), a small number of flyways can significantly improve performance, without the cost of building a fully non-oversubscribed network. ... We believe that 60 GHz flyways are an attractive choice because wireless devices simplify DC upgrades, as no wiring changes are needed. Furthermore, 60 GHz technology is likely to become inexpensive as it is commoditized by
consumer applications, while optical switches are not. ... indoor 60 GHz technologies such as 802.11ad target a short range of 10 meters, and 60 GHz links use highly directional antennas.*
Sounds interesting. But why not go for the comm laser?
Unless the Data Centre is in a Faraday Cage wouldn't wireless hacking be a problem?
Just a thought...
My working experience tells me that having a datacenter inside a Faraday cage is a VERY BAD idea.
¿Why? Well, just assume that you are the administrator of critical systems inside a steel reinforced, secure entry system with fingerprint, video, photos, secure card, daily code, one steel door, another security glass conductive door..
Ok, now you have a problem with the hardware of the system, or you might need to physically reboot it. No problem, it is in your own building: you go and get to solve the problem.
If you have no problems you can't/don't know how to fix, ok, but if you have problems, you can't phone your buddies so they remotelly help you: no phone. Ok, you might have one in the control room, but now you are maybe 100m away from your server.
If you send an operator, he will have the very same problem: no communication. So if problems arise, you might have him 80% or more of the time walking instead of solving the problem.. and you will go to the server yourself if you are not far from it.
Let's say that you want to have online help with your laptop near the server, using 3G for internet.. no luck, remember, faraday cage. the same goes for your Gold harware service technician...
I have worked in those conditions, and while you CAN still work, it is a less than desirable method of solving problems.
As for having high frequency transmissions inside a room filled with servers.. great, more RFI.. what you need is using optic cable to get rid of noise, not putting radiating antennas!! you might increase the error rates of your systems.. and that is really not desairable!! Also, if you use radio frequency connections, these connections have a greater error rate than cable ones, and retries might introduce jitter, and this is a big problem if you use those connections to manage high availability/replication/clusters... your systems might stop for sync!!
The solution is clear: you need an on rack switch, they are sold for a very good reason. Also, machines designed for housing blades also usually come with an internal switch for the very same reasons.. even if it is sometimes a bit underpowered.
You do realize...
That it takes a _lot_ more effort to build radio equipment which efficiently uses the radio spectrum than to simply build a faster Ethernet switch, do you?
And if you have more or less static traffic requirements, you could even go a step further by installing direct links between the server. Either by laying direct cables, or by using "hardwired" switches which don't care about the data and just connect 2 ports.
Cook your toast
Between two wireless network nodes.
Re: Cook your toast
I think that you'll find that :
1 - At 60GHz, the toast will be left mostly untouched by the RF energy.
2 - In the case of 2.4GHz (uW oven) the toat will be hot, but flappy. Best use the link to heat ramen.
Fire Icon, as IR is the only way to really do toasts.
You win the internets today for use of the word "flappy".
It would be cheaper to use a 10GB ethernet switch, additional 10GB ports and link the servers together.. with a dedicated link.
Anyway, if you are replicating, that is the way they should have been connected in the first place..
Gigabit Ethernet ?
Isn't that over copper ?
Doesn't fibre give largely superior speeds, with additional immunity from electromagnetic interference ?
Why are we even talking about wireless ?
Glad to see that everyone else
Has twigged that this is nonsense - either as an idea or as a piece of technical reporting.
"60 GHz links use highly directional antennas.*
Sounds interesting. But why not go for the comm laser?"
This whole concept has more holes than a Swiss cheese.
The idea that you will be able to get a reliable stable co-channel-free error-free 60GHz path between two arbitrary cabinets in a datacentre where locations of things change relatively frequently is not one of this idea's biggest problems, whether or not the antennae are directional, but it is a problem.
Anyone who's tried to have a 2.4GHz videosender coexist with 2.4GHz WiFi on the same block will understand this from practical experience.
Use a laser, poke it down a piece of glass string between cabinets, near-infinite bandwidth instantly available (but have your credit cards at the ready).
Someone has something to sell
This sounds very much like someone has some interesting 60GHz technology and is struggling to "monetise" it.
It does not sound like someone has found a sensible solution to a real problem.
Looks like selling
That technology might be interesting.
The problem I see is that it is 60Ghz. Yes, it will have plenty of BW IF you are able to encode/transmit/decode at the right speed and with adecuate reliability.. some 25 times more than wify.. but it will also reach some 25 less distance.. so your 100 m with wify (unobstructed) gets to be 4m.
Humm, how can we sell this? Reply: how big is a rack? no more that 3m. Great, we have a market here...
Only thing is, for a puby 4m inside a datacenter, I perefer to use a CABLE.
If you start using high power hotspots inside office buildings and high gain amplifiers inside those hotspots, I DO see as a substitute of lan cables.. but you will have problems:
-Wify is already king.
-IP Phones. Powered by PoE.. you NEED the cable, and if you already have it.. why use wireless?
Isn't it obvious?
That's a lot of effort...
... to cater to a bunch of cheapskates that didn't bother with a couple spare fibres 'cross the racks. Suddenly the appearance of that marketeering company makes sense. Leveraging movable solutionizing at its finest. The only real lesson in there is that it doesn't pay to apply client-server oversubscription rules of thumb to "cloud". It eats more bandwidth.
60GHz wireless in the datacentre gotta be wonderful. Can't wait to find out what boxes turn out to be insufficiently shielded.
Wireless bandwidth is finite, it would become very easy to saturate the airwaves inside the DC, and all those metal racks and cables would likely cause interference too.
Instead, the costs of simply laying more cable between racks is pretty trivial, lay another bundle of fibre and run more 10G trunks
Some good points raised here, particularly around bandwidth, cost, and security concerns.
Here's one more point - wireless networks can be seriously degraded by EM radiation. A datacenter is one of the most challenging RF/EM environments that one could possibly imagine. I would be very surprised if you could get consistently reliable connections, without frame drops. Think of all the EM interference!
Finally, there is constantly a drive within the datacenter for [consistent] low latency. Wireless technologies have higher latency than wired equivalents. Is this technology going to actually be useable?
Up there the main problem is fluorescent lamps. Certainly no problem in a modern "lights out" data center. ;)
Latency actually doesn't need to be a problem, given the right protocols that could work. Just think of directed radio links. An analogue video link needs to have a constant latency which must not vary more than a few nanoseconds, for example. And this can be reached since the 1960s.
Again that's not the problem, the problems are much more obvious as pointed out by many.
Point to Point radio at 60 GHz
There's another problem. Although it's easy to have directional antennas at 60 GHz, that won't help you inside of buildings. Essentially the signal would be reflected on just about any surface, particularly walls, but also even the antenna of the other side. So the more "links" you have the more "background noise" those will have to deal with. At best you'll end up with something like free space optics... at worst it won't work at all.
Just try it out yourself. Enter a dark room and turn on a laser pointer. Suddenly your whole room is bright. (to some degree)
Seems like visible light could work better, and BTW, why do we need to hide it in fiber anyway?
While I feel that the need for such measures should be mostly easily avoided by proper network design and implementation, and most times this sort of issue could best be resolved by stringing some more fiber, I also suspect that there is some niche out there for a 60GHZ wireless implementation. However, my thoughts went immediately towards using visible light (or thereabout) Comm type lasers (as mentioned earlier) which should offer greater capacity and avoid most of the challenges of 60GHZ ( capacity, security, interference, etc), and, with a bit of motor control offer serious flexibility and things like "self configuration" (and kill stray bugs - insects :). In fact, could such technology replace actual fiber? If the data is being carried by light and we have short haul line of sight access, why not just leave out the fiber? It would simplify installations enormously if we didn't have to string data cables
Why stop at stray bugs and insects? Get the human interlopers, too!