Slashdot Mirror
Google Router Rumors
An anonymous reader writes "There's a new rumor that Google is developing its own router. The company won't comment on the story, but it's been in the hardware business for a while and expanded its presence with Android. If Larry Ellison can go halvsies with HP on a server, then Eric Schmidt should certainly be able to make Cisco nervous."
...to procrastinate on the CCNA test.
THL phish sticks
All I need now is google underwear that twitters for me with real time gps tracking so I know where I've been.
I seem to remember rumors of them building their own insane (10 GbE) hardware switches. And I don't think that's hard to imagine as nothing on the market at the time could possibly meet their needs.
... like many claims they could not be operating on TCP/IP stacks on the inside. Because it's such a resource hog in some respects but that's irrelevant--I'm certain they can apply some of their ideas universally. I would put my money on them being the leader in research on networks and network theory ... probably past Cisco even (although behind the NSA as no one's ever sure about those guys). I feel that networking is so closely tied to their bread and butter search application that they should be dumping huge R&D into that field. I can't offer proof but it certainly makes sense to me.
Of course, there's a lot of questions that remain to be answered
And all I can say is that it's about time someone put pressure on the home & enterprise networking hardware companies. What a stagnant squabbling market that has become.
My work here is dung.
I hope they include sensible and up-to-date standards and protocols. I'm thinking about the possibilities of the interface of the tomato firmware and importantly, inclusion of ipv6 support. If we want this to happen in this generation we need to get software support on at least basic networking devices(thinking of routers and OSes).
Knowledge is power. Knowledge shared is power lost.
It seems likely to me that since Google is full of really smart people who seem to have a touch of the NIH syndrome, it probably isn't surprising that they wanted to develop their own routers from scratch instead of paying through the nose for Cisco or Juniper devices, especially since they needed hundreds or thousands of them and really don't want to have to pay for support contracts. I'd see a Google router announcement as just a productization of something they already use internally, just like Protocol Buffers.
The problem is that Google develops tech internally that is extremely good at solving their problems, but they don't always apply well outside of Google. Protocol Buffers aren't exactly obsoleting XML and from all indications they probably never will. The Google router will probably be super fast and simple, but lack a whole bunch of the more obscure features. The problem is that there's someone out there for each one of those obscure features, and if you don't support it your product won't even make it in the door. This is a problem Juniper runs into a lot, they have good and fast hardware, but the only thing it does is route.
In fact the article points out that Google's router is most likely to compete directly with Juniper instead of Cisco.
I read the internet for the articles.
They'll be 100% on the up and up WRT implementing standards compliance, and will release every last detail as open source, no-strings-attached goodness for the world to use. Such an act would be a giant cudgel that they could use against arguments that they're embracing proprietary tactics. They should do for routers what Android is trying to do for phones.
Forgive my spelling from time to time. I'm often posting during short breaks.
Presumably the people that would buy stuff just because it was made by Google are not a major demographic. So Google will need to do something to
1) raise the barrier to entry, no point issuing a device that anyone could make with Linux and a '386. Also, many cisco routers (eg. the 1800 series) genuinely represent value for money.
2) Provide good quality support.
So to raise the barrier to entry, it has to be a pretty special product, maybe doing the most useful 80% of what a cisco does flawlessly and improving upon cisco in come other areas (ones I can think off of the top of my head are ease of deployment and virtualization (vrf)).
The other reason people insist on Cisco, even when there are other cheaper options, is that they believe Cisco support their product well with training and technical support. This in my experience is an illusion. By and large the Cisco TAC is awful and maintaining certification is expensive and time consuming and the training materials are riddled with misprints, bugs and corporate "best practices" that are self-serving to Cisco.
So Google have a huge hill to climb, but I'm sure that it can be done in the space of a couple of years.
Nullius in verba
TFA says that Juniper is doomed because Google is getting ready to switch to their own in-house brand of routers. I find this difficult to believe for several reasons. One is that even if Google is Juniper's biggest customer, one customer does not a demise make -- Juniper has many other customers, including the entire UUnet (MCI, WorldCom, Verizon Business, whatever they're calling themselves this year) backbone. But there are far more practical reasons. Routers contain a lot of specialized hardware designed for rapid switching of packets. Google may have a lot of smart people working for them, but they certainly don't have the resources on board to design and build all of those ASIC's and other custom hardware, and it doesn't really make sense for them to get into that business during a recession just for an in-house project. (And no, don't give me that line about how a fast enough server with multiple Ethernet cards can substitute for even a mid-grade Cisco or Juniper. I manage a data center network and know the numbers. It can't even come close, no matter how good the software is, because a general purpose computer has to forward every packet using software, while a real router only makes a routing decision once and then all the rest of the packets for that destination are switched in hardware at wire speed.)
Nothing to see here. Move along.
Tired of FB/Google censorship? Visit UNCENSORED!
My router works fine, and I don't have Google stealing all of my LAN packets and serving me ads.
A fucking grouter had better make me warm delicious waffles if they want me to buy it. Even then, I'd only use it to make waffles.
And now I'm off to amazon to look for a waffle maker.
I really hope they throw in IPv6. There are no consumer-level routers available with IPv6 support; it's been driving me crazy. Everyone will probably be forced to buy new routers in a few years anyway.
With that said, I think Google is probably developing a router for their own in-house use. I have doubts this will actually hit the consumer market.
google.slashdot
Of course Google would not waste time developing their own ASICs. Companies like Marvell, Broadcom, and Dune offer plenty to choose from, and companies such as FDRY and JNPR already use these to build their own offerings.
It only makes sense for Google to use the building blocks to make a device that meets their specific needs.
1. it's open source
2. it's open source
3. it's open source
and probably some other reasons too.
The Apple Airport Express and Airport Extreme routers support IPv6, although there's a bug in the latest firmware for doing configured tunnels.
This
I have this mental image of a case with wide blue, red, yellow, blue, green then red stripes as well as similarly colored network cables, ethernet jacks, lights and buttons....
BARF!!!
Oh, I'm sure it'll work great - but hide that bitch in the rear of your rack space, that's for sure.
It's interesting that Apple OSX has supported IPv6 for a while (probably a side-effect from using BSD) and Apple routers (Airport Extreme) supports IPv6 and (if I remember the specs right) tunneling IPv6 over IPv4 out of the box and enabled.
While that does not represent the vast majority of the computers/home routers in use, this does show that some companies are trying to start the trend.
DEMETRIUS: Villain, what hast thou done?
AARON: Villain, I have done thy mother.
Shakespeare invents 'your mom'
Everybody seems to be assuming that these new routers will be for sale. That's obviously not going to happen — there just isn't room in the marketplace for a new player, even if that player is Google. Breaking into a new hardware marketplace is hard. You have to develop sales channels, create a hardware support organization, set up an operations organization to manage production, etc. etc.
I know about these things because for the last couple of years my job has been to document some of Sun's hardware products. Before that I mostly documented software, and the shear complexity of designing, building, distributing, selling and supporting actual physical products still boggles my mind. At product team meetings I sometimes feel at sea, even though the technical concepts I have to deal with are actually much simpler than those I faced when I was on software product teams. The logistics are just mind boggling.
Google isn't set up to be "in the hardware business". They make their own servers because there are no manufacturers that are able to meet their specialized needs. Now they seem to have decided that their routers also require specialized in-house designs. They haven't tried to sell these servers to other companies, and they won't try to sell their routers. Even if they could hope to compete, it would mean building up the kind of technical bureaucracy that Google's top echelon has no interest in managing.
Hell, they don't really have a proper bureaucracy for the much simpler job of creating and distributing their software products. If they actually charged money for most of them, they'd be trouble.
And Android? How does Android count as being "in the hardware business"? Is Google selling a cell phone I haven't heard about?
Or Google could buy Juniper. Let the rumor drive down the stock and pick them up at fire sale prices.
See my blog http://ilovecookes.blogspot.com/ for light hearted technical information.
Post is exactly right. The ASICs are already out there and in use by pretty much everyone for their COTS routers.
When one gets into the carrier-scale equipment I don't have a clue how that stuff goes. But I've seen enough low-end ( $10,000) routers taken apart to know that AC's comments are accurate.
http://www.maxineudall.com/2010/02/should-economists-be-sued-for-malpractice.html
Am I the only one who read this and thought, "Hmmm, it must be time for Google to renew their support contracts with Juniper.".
"leak" a rumor about no longer needing Juniper, and watch juniper lower their support rates.
"...even if Google is Juniper's biggest customer, one customer does not a demise make..."
That really depends. For smart companies, they've sufficiently diversified their client base such that the loss of one will hurt but not cripple. Some clients, however, just become so damn big and a company simply can't get enough other clients or the increase the volume from the other existing clients high enough to balance against that one mega-client. Once one client represents a massive percentage of your revenue and the loss of that client would force you into immediate emergency restructuring in the hopes of survival, then yes, one client a demise can potentially make.
The questions really are: how many different types of ASICs and boards are in those routers plus how many of the ASICs cannot be replaced with FPGAs and how many of the different board types cannot be rationalized to a smaller number of types? Remember that Google probably doesn't need the level of flexibility offered across Juniper's product range. It is clear that Google already has expertise in chip design -- it's not hard to find board design expertise (either in-house or outsourced).
The real "Libtards" are the Libertarians!
Just like with the 10G switches, this has all the earmarks of something for purely internal use rather than something they're planning to sell. That means their current vendor, which is Juniper according to TFA, loses Google as a customer, but that's about it.
If anything, Cisco should be happy that their competitor is losing business.
Why go bringing CISCO into this. Apart from creating products people want to use (gmail, search, etc..) google has two main focuses: building a back end able to efficiently run those applications and ensuring the consumer has easy access to those services.
Android and google's actions in the spectrum market weren't made just to fuck around with products outside their core competencies. They were strategic moves made to ensure that customers on mobile devices didn't end up directed away from google products by someone controlling the network or providing the handset.
Similarly google isn't about to start competing in the router market just for kicks. It's outside of their core competencies and the potential for profit simply wouldn't justify the resource expenditure.
Likely google is working on a custom router to help make their backend more efficient. To take an educated guess I would imagine that they want to build in intelligent load balancing into their routers. In other words have the routers maintain information about where certain kinds of data live and/or what machines are heavily loaded and then intelligently send requests for computations to lightly loaded nodes near the data. They might also want to simply build in custom handling of packets for things like GFS.
Not only will google not bother to compete in the router market but I suspect they won't even allow the technology they use for this to escape the company. After all most of the people who would benefit from this kind of optimization are their direct competitors.
If you liked this thought maybe you would find my blog nice too:
It's an awful summary. Google isn't dumb enough to go compete in the router market. They are likely creating optimized routers to service their own backend.
Don't you remember this was the same thing that happened when information on GFS leaked or the custom OS versions they use in their data center. People hyped it up as if google was going to take on MS in the OS arena.
If you liked this thought maybe you would find my blog nice too:
I've often heard this referred to as the Wal-Mart effect. Once Wal-Mart distributes your product nationally, they basically own you. Because once you ramp up production to meet Wal-Mart needs, you can't just scale back down if they drop you... and they can and will drop you if you do not behave.
Check out my lame java blog at www.javachopshop.com
This is old news and was announced almost two years ago http://www.google.com/tisp/
I can't believe nobody has made mention of Vyatta. It's an excellent appliance-like distro based on, I believe, Debian.
It's not mentioned because it's not even remotely relevant to the discussion.
All the bells and whistles you'd expect from a high-end device at a fraction (by which I mean ~1/3) of the cost relative to a Cisco purchase.
Including bells and whistles like custom ASICs and switching fabrics? Oh, wait, it doesn't have those. Nothing about Vyatta is "high-end." It is, however, a viable alternative at the very low-end.
"Nature doesn't care how smart you are. You can still be wrong." - Richard Feynman
The lines between software and hardware are actually really blurry. Most NICs, for example, have hardware which assists in manipulating packets--anything from simply managing the checksums to VLAN tagging. Some cards even come with prioritization in the ASIC. Then you get highly programmable NICs which basically include an FPGA and a programming interface. With these, you can implement a somewhat arbitrary portion of the TCP/IP stack in the FPGA.
"But it's still softare!" you may cry. Well, maybe. But that's the point. The line between software and hardware is wide and blurry these days which, incidentally, is part of the reason why we have binary blogs for wireless drivers in the Linux kernel (they're basically firmware for the cards which the OS loads on boot.)
So saying "the software level" really just doesn't make sense. The layers in the OSI model don't distinguish between hardware and software--in fact, software isn't really mentioned except in layer 7 (the application layer.)
Steve Yegge (Google employee) hinted about this in June 2007. (He said he had to write a new parser for it.) Look at point number 5 here: http://steve-yegge.blogspot.com/2007/06/rich-programmer-food.html